1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019, 2020 Mellanox Technologies LTD. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/accel_engine.h" 35 #include "spdk/stdinc.h" 36 #include "spdk/crc32.h" 37 #include "spdk/endian.h" 38 #include "spdk/assert.h" 39 #include "spdk/thread.h" 40 #include "spdk/nvmf_transport.h" 41 #include "spdk/string.h" 42 #include "spdk/trace.h" 43 #include "spdk/util.h" 44 #include "spdk/log.h" 45 46 #include "spdk_internal/assert.h" 47 #include "spdk_internal/nvme_tcp.h" 48 #include "spdk_internal/sock.h" 49 50 #include "nvmf_internal.h" 51 52 #define NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME 16 53 #define SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY 16 54 #define SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY 0 55 #define SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM 32 56 #define SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION true 57 58 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp; 59 60 /* spdk nvmf related structure */ 61 enum spdk_nvmf_tcp_req_state { 62 63 /* The request is not currently in use */ 64 TCP_REQUEST_STATE_FREE = 0, 65 66 /* Initial state when request first received */ 67 TCP_REQUEST_STATE_NEW, 68 69 /* The request is queued until a data buffer is available. */ 70 TCP_REQUEST_STATE_NEED_BUFFER, 71 72 /* The request is currently transferring data from the host to the controller. */ 73 TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 74 75 /* The request is waiting for the R2T send acknowledgement. */ 76 TCP_REQUEST_STATE_AWAITING_R2T_ACK, 77 78 /* The request is ready to execute at the block device */ 79 TCP_REQUEST_STATE_READY_TO_EXECUTE, 80 81 /* The request is currently executing at the block device */ 82 TCP_REQUEST_STATE_EXECUTING, 83 84 /* The request finished executing at the block device */ 85 TCP_REQUEST_STATE_EXECUTED, 86 87 /* The request is ready to send a completion */ 88 TCP_REQUEST_STATE_READY_TO_COMPLETE, 89 90 /* The request is currently transferring final pdus from the controller to the host. */ 91 TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 92 93 /* The request completed and can be marked free. */ 94 TCP_REQUEST_STATE_COMPLETED, 95 96 /* Terminator */ 97 TCP_REQUEST_NUM_STATES, 98 }; 99 100 static const char *spdk_nvmf_tcp_term_req_fes_str[] = { 101 "Invalid PDU Header Field", 102 "PDU Sequence Error", 103 "Header Digiest Error", 104 "Data Transfer Out of Range", 105 "R2T Limit Exceeded", 106 "Unsupported parameter", 107 }; 108 109 #define OBJECT_NVMF_TCP_IO 0x80 110 111 #define TRACE_GROUP_NVMF_TCP 0x5 112 #define TRACE_TCP_REQUEST_STATE_NEW SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x0) 113 #define TRACE_TCP_REQUEST_STATE_NEED_BUFFER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x1) 114 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x2) 115 #define TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x3) 116 #define TRACE_TCP_REQUEST_STATE_EXECUTING SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x4) 117 #define TRACE_TCP_REQUEST_STATE_EXECUTED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x5) 118 #define TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x6) 119 #define TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x7) 120 #define TRACE_TCP_REQUEST_STATE_COMPLETED SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x8) 121 #define TRACE_TCP_FLUSH_WRITEBUF_START SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0x9) 122 #define TRACE_TCP_FLUSH_WRITEBUF_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xA) 123 #define TRACE_TCP_READ_FROM_SOCKET_DONE SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xB) 124 #define TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK SPDK_TPOINT_ID(TRACE_GROUP_NVMF_TCP, 0xC) 125 126 SPDK_TRACE_REGISTER_FN(nvmf_tcp_trace, "nvmf_tcp", TRACE_GROUP_NVMF_TCP) 127 { 128 spdk_trace_register_object(OBJECT_NVMF_TCP_IO, 'r'); 129 spdk_trace_register_description("TCP_REQ_NEW", 130 TRACE_TCP_REQUEST_STATE_NEW, 131 OWNER_NONE, OBJECT_NVMF_TCP_IO, 1, 1, ""); 132 spdk_trace_register_description("TCP_REQ_NEED_BUFFER", 133 TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 134 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 135 spdk_trace_register_description("TCP_REQ_TX_H_TO_C", 136 TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 137 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 138 spdk_trace_register_description("TCP_REQ_RDY_TO_EXECUTE", 139 TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 140 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 141 spdk_trace_register_description("TCP_REQ_EXECUTING", 142 TRACE_TCP_REQUEST_STATE_EXECUTING, 143 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 144 spdk_trace_register_description("TCP_REQ_EXECUTED", 145 TRACE_TCP_REQUEST_STATE_EXECUTED, 146 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 147 spdk_trace_register_description("TCP_REQ_RDY_TO_COMPLETE", 148 TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 149 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 150 spdk_trace_register_description("TCP_REQ_TRANSFER_C2H", 151 TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 152 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 153 spdk_trace_register_description("TCP_REQ_COMPLETED", 154 TRACE_TCP_REQUEST_STATE_COMPLETED, 155 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 156 spdk_trace_register_description("TCP_WRITE_START", 157 TRACE_TCP_FLUSH_WRITEBUF_START, 158 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 159 spdk_trace_register_description("TCP_WRITE_DONE", 160 TRACE_TCP_FLUSH_WRITEBUF_DONE, 161 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 162 spdk_trace_register_description("TCP_READ_DONE", 163 TRACE_TCP_READ_FROM_SOCKET_DONE, 164 OWNER_NONE, OBJECT_NONE, 0, 0, ""); 165 spdk_trace_register_description("TCP_REQ_AWAIT_R2T_ACK", 166 TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 167 OWNER_NONE, OBJECT_NVMF_TCP_IO, 0, 1, ""); 168 } 169 170 struct spdk_nvmf_tcp_req { 171 struct spdk_nvmf_request req; 172 struct spdk_nvme_cpl rsp; 173 struct spdk_nvme_cmd cmd; 174 175 /* A PDU that can be used for sending responses. This is 176 * not the incoming PDU! */ 177 struct nvme_tcp_pdu *pdu; 178 179 /* In-capsule data buffer */ 180 uint8_t *buf; 181 /* 182 * The PDU for a request may be used multiple times in serial over 183 * the request's lifetime. For example, first to send an R2T, then 184 * to send a completion. To catch mistakes where the PDU is used 185 * twice at the same time, add a debug flag here for init/fini. 186 */ 187 bool pdu_in_use; 188 bool has_incapsule_data; 189 190 /* transfer_tag */ 191 uint16_t ttag; 192 193 enum spdk_nvmf_tcp_req_state state; 194 195 /* 196 * h2c_offset is used when we receive the h2c_data PDU. 197 */ 198 uint32_t h2c_offset; 199 200 STAILQ_ENTRY(spdk_nvmf_tcp_req) link; 201 TAILQ_ENTRY(spdk_nvmf_tcp_req) state_link; 202 }; 203 204 struct spdk_nvmf_tcp_qpair { 205 struct spdk_nvmf_qpair qpair; 206 struct spdk_nvmf_tcp_poll_group *group; 207 struct spdk_sock *sock; 208 209 enum nvme_tcp_pdu_recv_state recv_state; 210 enum nvme_tcp_qpair_state state; 211 212 /* PDU being actively received */ 213 struct nvme_tcp_pdu pdu_in_progress; 214 215 /* Queues to track the requests in all states */ 216 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_working_queue; 217 TAILQ_HEAD(, spdk_nvmf_tcp_req) tcp_req_free_queue; 218 219 /* Number of requests in each state */ 220 uint32_t state_cntr[TCP_REQUEST_NUM_STATES]; 221 222 uint8_t cpda; 223 224 bool host_hdgst_enable; 225 bool host_ddgst_enable; 226 227 /* This is a spare PDU used for sending special management 228 * operations. Primarily, this is used for the initial 229 * connection response and c2h termination request. */ 230 struct nvme_tcp_pdu *mgmt_pdu; 231 232 /* Arrays of in-capsule buffers, requests, and pdus. 233 * Each array is 'resource_count' number of elements */ 234 void *bufs; 235 struct spdk_nvmf_tcp_req *reqs; 236 struct nvme_tcp_pdu *pdus; 237 uint32_t resource_count; 238 uint32_t recv_buf_size; 239 240 struct spdk_nvmf_tcp_port *port; 241 242 /* IP address */ 243 char initiator_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 244 char target_addr[SPDK_NVMF_TRADDR_MAX_LEN]; 245 246 /* IP port */ 247 uint16_t initiator_port; 248 uint16_t target_port; 249 250 /* Timer used to destroy qpair after detecting transport error issue if initiator does 251 * not close the connection. 252 */ 253 struct spdk_poller *timeout_poller; 254 255 struct spdk_io_channel *accel_channel; 256 257 TAILQ_ENTRY(spdk_nvmf_tcp_qpair) link; 258 }; 259 260 struct spdk_nvmf_tcp_control_msg { 261 STAILQ_ENTRY(spdk_nvmf_tcp_control_msg) link; 262 }; 263 264 struct spdk_nvmf_tcp_control_msg_list { 265 void *msg_buf; 266 STAILQ_HEAD(, spdk_nvmf_tcp_control_msg) free_msgs; 267 }; 268 269 struct spdk_nvmf_tcp_poll_group { 270 struct spdk_nvmf_transport_poll_group group; 271 struct spdk_sock_group *sock_group; 272 273 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) qpairs; 274 TAILQ_HEAD(, spdk_nvmf_tcp_qpair) await_req; 275 276 struct spdk_nvmf_tcp_control_msg_list *control_msg_list; 277 }; 278 279 struct spdk_nvmf_tcp_port { 280 const struct spdk_nvme_transport_id *trid; 281 struct spdk_sock *listen_sock; 282 TAILQ_ENTRY(spdk_nvmf_tcp_port) link; 283 }; 284 285 struct tcp_transport_opts { 286 bool c2h_success; 287 uint16_t control_msg_num; 288 uint32_t sock_priority; 289 }; 290 291 struct spdk_nvmf_tcp_transport { 292 struct spdk_nvmf_transport transport; 293 struct tcp_transport_opts tcp_opts; 294 295 pthread_mutex_t lock; 296 297 TAILQ_HEAD(, spdk_nvmf_tcp_port) ports; 298 }; 299 300 static const struct spdk_json_object_decoder tcp_transport_opts_decoder[] = { 301 { 302 "c2h_success", offsetof(struct tcp_transport_opts, c2h_success), 303 spdk_json_decode_bool, true 304 }, 305 { 306 "control_msg_num", offsetof(struct tcp_transport_opts, control_msg_num), 307 spdk_json_decode_uint16, true 308 }, 309 { 310 "sock_priority", offsetof(struct tcp_transport_opts, sock_priority), 311 spdk_json_decode_uint32, true 312 }, 313 }; 314 315 static bool nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 316 struct spdk_nvmf_tcp_req *tcp_req); 317 static void nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group); 318 319 static void 320 nvmf_tcp_req_set_state(struct spdk_nvmf_tcp_req *tcp_req, 321 enum spdk_nvmf_tcp_req_state state) 322 { 323 struct spdk_nvmf_qpair *qpair; 324 struct spdk_nvmf_tcp_qpair *tqpair; 325 326 qpair = tcp_req->req.qpair; 327 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 328 329 assert(tqpair->state_cntr[tcp_req->state] > 0); 330 tqpair->state_cntr[tcp_req->state]--; 331 tqpair->state_cntr[state]++; 332 333 tcp_req->state = state; 334 } 335 336 static inline struct nvme_tcp_pdu * 337 nvmf_tcp_req_pdu_init(struct spdk_nvmf_tcp_req *tcp_req) 338 { 339 assert(tcp_req->pdu_in_use == false); 340 tcp_req->pdu_in_use = true; 341 342 memset(tcp_req->pdu, 0, sizeof(*tcp_req->pdu)); 343 tcp_req->pdu->qpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 344 345 return tcp_req->pdu; 346 } 347 348 static inline void 349 nvmf_tcp_req_pdu_fini(struct spdk_nvmf_tcp_req *tcp_req) 350 { 351 tcp_req->pdu_in_use = false; 352 } 353 354 static struct spdk_nvmf_tcp_req * 355 nvmf_tcp_req_get(struct spdk_nvmf_tcp_qpair *tqpair) 356 { 357 struct spdk_nvmf_tcp_req *tcp_req; 358 359 tcp_req = TAILQ_FIRST(&tqpair->tcp_req_free_queue); 360 if (!tcp_req) { 361 return NULL; 362 } 363 364 memset(&tcp_req->rsp, 0, sizeof(tcp_req->rsp)); 365 tcp_req->h2c_offset = 0; 366 tcp_req->has_incapsule_data = false; 367 tcp_req->req.dif.dif_insert_or_strip = false; 368 369 TAILQ_REMOVE(&tqpair->tcp_req_free_queue, tcp_req, state_link); 370 TAILQ_INSERT_TAIL(&tqpair->tcp_req_working_queue, tcp_req, state_link); 371 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEW); 372 return tcp_req; 373 } 374 375 static inline void 376 nvmf_tcp_req_put(struct spdk_nvmf_tcp_qpair *tqpair, struct spdk_nvmf_tcp_req *tcp_req) 377 { 378 TAILQ_REMOVE(&tqpair->tcp_req_working_queue, tcp_req, state_link); 379 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 380 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_FREE); 381 } 382 383 static void 384 nvmf_tcp_request_free(void *cb_arg) 385 { 386 struct spdk_nvmf_tcp_transport *ttransport; 387 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 388 389 assert(tcp_req != NULL); 390 391 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req=%p will be freed\n", tcp_req); 392 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 393 struct spdk_nvmf_tcp_transport, transport); 394 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 395 nvmf_tcp_req_process(ttransport, tcp_req); 396 } 397 398 static int 399 nvmf_tcp_req_free(struct spdk_nvmf_request *req) 400 { 401 struct spdk_nvmf_tcp_req *tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 402 403 nvmf_tcp_request_free(tcp_req); 404 405 return 0; 406 } 407 408 static void 409 nvmf_tcp_drain_state_queue(struct spdk_nvmf_tcp_qpair *tqpair, 410 enum spdk_nvmf_tcp_req_state state) 411 { 412 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 413 414 assert(state != TCP_REQUEST_STATE_FREE); 415 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 416 if (state == tcp_req->state) { 417 nvmf_tcp_request_free(tcp_req); 418 } 419 } 420 } 421 422 static void 423 nvmf_tcp_cleanup_all_states(struct spdk_nvmf_tcp_qpair *tqpair) 424 { 425 struct spdk_nvmf_tcp_req *tcp_req, *req_tmp; 426 427 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 428 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEW); 429 430 /* Wipe the requests waiting for buffer from the global list */ 431 TAILQ_FOREACH_SAFE(tcp_req, &tqpair->tcp_req_working_queue, state_link, req_tmp) { 432 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 433 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, &tcp_req->req, 434 spdk_nvmf_request, buf_link); 435 } 436 } 437 438 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_NEED_BUFFER); 439 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_EXECUTING); 440 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 441 nvmf_tcp_drain_state_queue(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 442 } 443 444 static void 445 nvmf_tcp_dump_qpair_req_contents(struct spdk_nvmf_tcp_qpair *tqpair) 446 { 447 int i; 448 struct spdk_nvmf_tcp_req *tcp_req; 449 450 SPDK_ERRLOG("Dumping contents of queue pair (QID %d)\n", tqpair->qpair.qid); 451 for (i = 1; i < TCP_REQUEST_NUM_STATES; i++) { 452 SPDK_ERRLOG("\tNum of requests in state[%d] = %u\n", i, tqpair->state_cntr[i]); 453 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 454 if ((int)tcp_req->state == i) { 455 SPDK_ERRLOG("\t\tRequest Data From Pool: %d\n", tcp_req->req.data_from_pool); 456 SPDK_ERRLOG("\t\tRequest opcode: %d\n", tcp_req->req.cmd->nvmf_cmd.opcode); 457 } 458 } 459 } 460 } 461 462 static void 463 nvmf_tcp_qpair_destroy(struct spdk_nvmf_tcp_qpair *tqpair) 464 { 465 int err = 0; 466 467 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 468 469 err = spdk_sock_close(&tqpair->sock); 470 assert(err == 0); 471 nvmf_tcp_cleanup_all_states(tqpair); 472 473 if (tqpair->state_cntr[TCP_REQUEST_STATE_FREE] != tqpair->resource_count) { 474 SPDK_ERRLOG("tqpair(%p) free tcp request num is %u but should be %u\n", tqpair, 475 tqpair->state_cntr[TCP_REQUEST_STATE_FREE], 476 tqpair->resource_count); 477 err++; 478 } 479 480 if (err > 0) { 481 nvmf_tcp_dump_qpair_req_contents(tqpair); 482 } 483 484 if (tqpair->accel_channel) { 485 spdk_put_io_channel(tqpair->accel_channel); 486 } 487 spdk_dma_free(tqpair->pdus); 488 free(tqpair->reqs); 489 spdk_free(tqpair->bufs); 490 free(tqpair); 491 SPDK_DEBUGLOG(nvmf_tcp, "Leave\n"); 492 } 493 494 static void 495 nvmf_tcp_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w) 496 { 497 struct spdk_nvmf_tcp_transport *ttransport; 498 assert(w != NULL); 499 500 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 501 spdk_json_write_named_bool(w, "c2h_success", ttransport->tcp_opts.c2h_success); 502 spdk_json_write_named_uint32(w, "sock_priority", ttransport->tcp_opts.sock_priority); 503 } 504 505 static int 506 nvmf_tcp_destroy(struct spdk_nvmf_transport *transport, 507 spdk_nvmf_transport_destroy_done_cb cb_fn, void *cb_arg) 508 { 509 struct spdk_nvmf_tcp_transport *ttransport; 510 511 assert(transport != NULL); 512 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 513 514 pthread_mutex_destroy(&ttransport->lock); 515 free(ttransport); 516 517 if (cb_fn) { 518 cb_fn(cb_arg); 519 } 520 return 0; 521 } 522 523 static struct spdk_nvmf_transport * 524 nvmf_tcp_create(struct spdk_nvmf_transport_opts *opts) 525 { 526 struct spdk_nvmf_tcp_transport *ttransport; 527 uint32_t sge_count; 528 uint32_t min_shared_buffers; 529 530 ttransport = calloc(1, sizeof(*ttransport)); 531 if (!ttransport) { 532 return NULL; 533 } 534 535 TAILQ_INIT(&ttransport->ports); 536 537 ttransport->transport.ops = &spdk_nvmf_transport_tcp; 538 539 ttransport->tcp_opts.c2h_success = SPDK_NVMF_TCP_DEFAULT_SUCCESS_OPTIMIZATION; 540 ttransport->tcp_opts.sock_priority = SPDK_NVMF_TCP_DEFAULT_SOCK_PRIORITY; 541 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 542 if (opts->transport_specific != NULL && 543 spdk_json_decode_object_relaxed(opts->transport_specific, tcp_transport_opts_decoder, 544 SPDK_COUNTOF(tcp_transport_opts_decoder), 545 &ttransport->tcp_opts)) { 546 SPDK_ERRLOG("spdk_json_decode_object_relaxed failed\n"); 547 free(ttransport); 548 return NULL; 549 } 550 551 SPDK_NOTICELOG("*** TCP Transport Init ***\n"); 552 553 SPDK_INFOLOG(nvmf_tcp, "*** TCP Transport Init ***\n" 554 " Transport opts: max_ioq_depth=%d, max_io_size=%d,\n" 555 " max_io_qpairs_per_ctrlr=%d, io_unit_size=%d,\n" 556 " in_capsule_data_size=%d, max_aq_depth=%d\n" 557 " num_shared_buffers=%d, c2h_success=%d,\n" 558 " dif_insert_or_strip=%d, sock_priority=%d\n" 559 " abort_timeout_sec=%d, control_msg_num=%hu\n", 560 opts->max_queue_depth, 561 opts->max_io_size, 562 opts->max_qpairs_per_ctrlr - 1, 563 opts->io_unit_size, 564 opts->in_capsule_data_size, 565 opts->max_aq_depth, 566 opts->num_shared_buffers, 567 ttransport->tcp_opts.c2h_success, 568 opts->dif_insert_or_strip, 569 ttransport->tcp_opts.sock_priority, 570 opts->abort_timeout_sec, 571 ttransport->tcp_opts.control_msg_num); 572 573 if (ttransport->tcp_opts.sock_priority > SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY) { 574 SPDK_ERRLOG("Unsupported socket_priority=%d, the current range is: 0 to %d\n" 575 "you can use man 7 socket to view the range of priority under SO_PRIORITY item\n", 576 ttransport->tcp_opts.sock_priority, SPDK_NVMF_TCP_DEFAULT_MAX_SOCK_PRIORITY); 577 free(ttransport); 578 return NULL; 579 } 580 581 if (ttransport->tcp_opts.control_msg_num == 0 && 582 opts->in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 583 SPDK_WARNLOG("TCP param control_msg_num can't be 0 if ICD is less than %u bytes. Using default value %u\n", 584 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM); 585 ttransport->tcp_opts.control_msg_num = SPDK_NVMF_TCP_DEFAULT_CONTROL_MSG_NUM; 586 } 587 588 /* I/O unit size cannot be larger than max I/O size */ 589 if (opts->io_unit_size > opts->max_io_size) { 590 opts->io_unit_size = opts->max_io_size; 591 } 592 593 sge_count = opts->max_io_size / opts->io_unit_size; 594 if (sge_count > SPDK_NVMF_MAX_SGL_ENTRIES) { 595 SPDK_ERRLOG("Unsupported IO Unit size specified, %d bytes\n", opts->io_unit_size); 596 free(ttransport); 597 return NULL; 598 } 599 600 min_shared_buffers = spdk_env_get_core_count() * opts->buf_cache_size; 601 if (min_shared_buffers > opts->num_shared_buffers) { 602 SPDK_ERRLOG("There are not enough buffers to satisfy" 603 "per-poll group caches for each thread. (%" PRIu32 ")" 604 "supplied. (%" PRIu32 ") required\n", opts->num_shared_buffers, min_shared_buffers); 605 SPDK_ERRLOG("Please specify a larger number of shared buffers\n"); 606 nvmf_tcp_destroy(&ttransport->transport, NULL, NULL); 607 return NULL; 608 } 609 610 pthread_mutex_init(&ttransport->lock, NULL); 611 612 return &ttransport->transport; 613 } 614 615 static int 616 nvmf_tcp_trsvcid_to_int(const char *trsvcid) 617 { 618 unsigned long long ull; 619 char *end = NULL; 620 621 ull = strtoull(trsvcid, &end, 10); 622 if (end == NULL || end == trsvcid || *end != '\0') { 623 return -1; 624 } 625 626 /* Valid TCP/IP port numbers are in [0, 65535] */ 627 if (ull > 65535) { 628 return -1; 629 } 630 631 return (int)ull; 632 } 633 634 /** 635 * Canonicalize a listen address trid. 636 */ 637 static int 638 nvmf_tcp_canon_listen_trid(struct spdk_nvme_transport_id *canon_trid, 639 const struct spdk_nvme_transport_id *trid) 640 { 641 int trsvcid_int; 642 643 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 644 if (trsvcid_int < 0) { 645 return -EINVAL; 646 } 647 648 memset(canon_trid, 0, sizeof(*canon_trid)); 649 spdk_nvme_trid_populate_transport(canon_trid, SPDK_NVME_TRANSPORT_TCP); 650 canon_trid->adrfam = trid->adrfam; 651 snprintf(canon_trid->traddr, sizeof(canon_trid->traddr), "%s", trid->traddr); 652 snprintf(canon_trid->trsvcid, sizeof(canon_trid->trsvcid), "%d", trsvcid_int); 653 654 return 0; 655 } 656 657 /** 658 * Find an existing listening port. 659 * 660 * Caller must hold ttransport->lock. 661 */ 662 static struct spdk_nvmf_tcp_port * 663 nvmf_tcp_find_port(struct spdk_nvmf_tcp_transport *ttransport, 664 const struct spdk_nvme_transport_id *trid) 665 { 666 struct spdk_nvme_transport_id canon_trid; 667 struct spdk_nvmf_tcp_port *port; 668 669 if (nvmf_tcp_canon_listen_trid(&canon_trid, trid) != 0) { 670 return NULL; 671 } 672 673 TAILQ_FOREACH(port, &ttransport->ports, link) { 674 if (spdk_nvme_transport_id_compare(&canon_trid, port->trid) == 0) { 675 return port; 676 } 677 } 678 679 return NULL; 680 } 681 682 static int 683 nvmf_tcp_listen(struct spdk_nvmf_transport *transport, const struct spdk_nvme_transport_id *trid, 684 struct spdk_nvmf_listen_opts *listen_opts) 685 { 686 struct spdk_nvmf_tcp_transport *ttransport; 687 struct spdk_nvmf_tcp_port *port; 688 int trsvcid_int; 689 uint8_t adrfam; 690 struct spdk_sock_opts opts; 691 692 if (!strlen(trid->trsvcid)) { 693 SPDK_ERRLOG("Service id is required\n"); 694 return -EINVAL; 695 } 696 697 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 698 699 trsvcid_int = nvmf_tcp_trsvcid_to_int(trid->trsvcid); 700 if (trsvcid_int < 0) { 701 SPDK_ERRLOG("Invalid trsvcid '%s'\n", trid->trsvcid); 702 return -EINVAL; 703 } 704 705 pthread_mutex_lock(&ttransport->lock); 706 port = calloc(1, sizeof(*port)); 707 if (!port) { 708 SPDK_ERRLOG("Port allocation failed\n"); 709 pthread_mutex_unlock(&ttransport->lock); 710 return -ENOMEM; 711 } 712 713 port->trid = trid; 714 opts.opts_size = sizeof(opts); 715 spdk_sock_get_default_opts(&opts); 716 opts.priority = ttransport->tcp_opts.sock_priority; 717 port->listen_sock = spdk_sock_listen_ext(trid->traddr, trsvcid_int, 718 NULL, &opts); 719 if (port->listen_sock == NULL) { 720 SPDK_ERRLOG("spdk_sock_listen(%s, %d) failed: %s (%d)\n", 721 trid->traddr, trsvcid_int, 722 spdk_strerror(errno), errno); 723 free(port); 724 pthread_mutex_unlock(&ttransport->lock); 725 return -errno; 726 } 727 728 if (spdk_sock_is_ipv4(port->listen_sock)) { 729 adrfam = SPDK_NVMF_ADRFAM_IPV4; 730 } else if (spdk_sock_is_ipv6(port->listen_sock)) { 731 adrfam = SPDK_NVMF_ADRFAM_IPV6; 732 } else { 733 SPDK_ERRLOG("Unhandled socket type\n"); 734 adrfam = 0; 735 } 736 737 if (adrfam != trid->adrfam) { 738 SPDK_ERRLOG("Socket address family mismatch\n"); 739 spdk_sock_close(&port->listen_sock); 740 free(port); 741 pthread_mutex_unlock(&ttransport->lock); 742 return -EINVAL; 743 } 744 745 SPDK_NOTICELOG("*** NVMe/TCP Target Listening on %s port %s ***\n", 746 trid->traddr, trid->trsvcid); 747 748 TAILQ_INSERT_TAIL(&ttransport->ports, port, link); 749 pthread_mutex_unlock(&ttransport->lock); 750 return 0; 751 } 752 753 static void 754 nvmf_tcp_stop_listen(struct spdk_nvmf_transport *transport, 755 const struct spdk_nvme_transport_id *trid) 756 { 757 struct spdk_nvmf_tcp_transport *ttransport; 758 struct spdk_nvmf_tcp_port *port; 759 760 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 761 762 SPDK_DEBUGLOG(nvmf_tcp, "Removing listen address %s port %s\n", 763 trid->traddr, trid->trsvcid); 764 765 pthread_mutex_lock(&ttransport->lock); 766 port = nvmf_tcp_find_port(ttransport, trid); 767 if (port) { 768 TAILQ_REMOVE(&ttransport->ports, port, link); 769 spdk_sock_close(&port->listen_sock); 770 free(port); 771 } 772 773 pthread_mutex_unlock(&ttransport->lock); 774 } 775 776 static void nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 777 enum nvme_tcp_pdu_recv_state state); 778 779 static void 780 nvmf_tcp_qpair_disconnect(struct spdk_nvmf_tcp_qpair *tqpair) 781 { 782 SPDK_DEBUGLOG(nvmf_tcp, "Disconnecting qpair %p\n", tqpair); 783 784 if (tqpair->state <= NVME_TCP_QPAIR_STATE_RUNNING) { 785 tqpair->state = NVME_TCP_QPAIR_STATE_EXITING; 786 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 787 spdk_poller_unregister(&tqpair->timeout_poller); 788 789 /* This will end up calling nvmf_tcp_close_qpair */ 790 spdk_nvmf_qpair_disconnect(&tqpair->qpair, NULL, NULL); 791 } 792 } 793 794 static void 795 _pdu_write_done(void *_pdu, int err) 796 { 797 struct nvme_tcp_pdu *pdu = _pdu; 798 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 799 800 if (err != 0) { 801 nvmf_tcp_qpair_disconnect(tqpair); 802 return; 803 } 804 805 assert(pdu->cb_fn != NULL); 806 pdu->cb_fn(pdu->cb_arg); 807 } 808 809 static void 810 _tcp_write_pdu(struct nvme_tcp_pdu *pdu) 811 { 812 uint32_t mapped_length = 0; 813 ssize_t rc; 814 struct spdk_nvmf_tcp_qpair *tqpair = pdu->qpair; 815 uint32_t crc32c; 816 817 /* Data Digest */ 818 if (pdu->data_len > 0 && g_nvme_tcp_ddgst[pdu->hdr.common.pdu_type] && tqpair->host_ddgst_enable) { 819 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 820 MAKE_DIGEST_WORD(pdu->data_digest, crc32c); 821 } 822 823 pdu->sock_req.iovcnt = nvme_tcp_build_iovs(pdu->iov, SPDK_COUNTOF(pdu->iov), pdu, 824 tqpair->host_hdgst_enable, tqpair->host_ddgst_enable, 825 &mapped_length); 826 pdu->sock_req.cb_fn = _pdu_write_done; 827 pdu->sock_req.cb_arg = pdu; 828 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_RESP || 829 pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ) { 830 rc = spdk_sock_writev(tqpair->sock, pdu->iov, pdu->sock_req.iovcnt); 831 if (rc == mapped_length) { 832 _pdu_write_done(pdu, 0); 833 } else { 834 SPDK_ERRLOG("IC_RESP or TERM_REQ could not write to socket.\n"); 835 _pdu_write_done(pdu, -1); 836 } 837 } else { 838 spdk_sock_writev_async(tqpair->sock, &pdu->sock_req); 839 } 840 } 841 842 static void 843 header_crc32_accel_done(void *cb_arg, int status) 844 { 845 struct nvme_tcp_pdu *pdu = cb_arg; 846 847 pdu->header_digest_crc32 = pdu->header_digest_crc32 ^ SPDK_CRC32C_XOR; 848 MAKE_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, pdu->header_digest_crc32); 849 if (spdk_unlikely(status)) { 850 SPDK_ERRLOG("Failed to finish the crc32 work\n"); 851 _pdu_write_done(pdu, status); 852 return; 853 } 854 855 _tcp_write_pdu(pdu); 856 } 857 858 static void 859 nvmf_tcp_qpair_write_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 860 struct nvme_tcp_pdu *pdu, 861 nvme_tcp_qpair_xfer_complete_cb cb_fn, 862 void *cb_arg) 863 { 864 int hlen; 865 866 assert(&tqpair->pdu_in_progress != pdu); 867 868 hlen = pdu->hdr.common.hlen; 869 pdu->cb_fn = cb_fn; 870 pdu->cb_arg = cb_arg; 871 pdu->qpair = tqpair; 872 873 /* Header Digest */ 874 if (g_nvme_tcp_hdgst[pdu->hdr.common.pdu_type] && tqpair->host_hdgst_enable) { 875 spdk_accel_submit_crc32c(tqpair->accel_channel, &pdu->header_digest_crc32, 876 &pdu->hdr.raw, 0, 877 hlen, header_crc32_accel_done, pdu); 878 return; 879 } 880 881 _tcp_write_pdu(pdu); 882 } 883 884 static int 885 nvmf_tcp_qpair_init_mem_resource(struct spdk_nvmf_tcp_qpair *tqpair) 886 { 887 uint32_t i; 888 struct spdk_nvmf_transport_opts *opts; 889 uint32_t in_capsule_data_size; 890 891 opts = &tqpair->qpair.transport->opts; 892 893 in_capsule_data_size = opts->in_capsule_data_size; 894 if (opts->dif_insert_or_strip) { 895 in_capsule_data_size = SPDK_BDEV_BUF_SIZE_WITH_MD(in_capsule_data_size); 896 } 897 898 tqpair->resource_count = opts->max_queue_depth; 899 900 tqpair->reqs = calloc(tqpair->resource_count, sizeof(*tqpair->reqs)); 901 if (!tqpair->reqs) { 902 SPDK_ERRLOG("Unable to allocate reqs on tqpair=%p\n", tqpair); 903 return -1; 904 } 905 906 if (in_capsule_data_size) { 907 tqpair->bufs = spdk_zmalloc(tqpair->resource_count * in_capsule_data_size, 0x1000, 908 NULL, SPDK_ENV_LCORE_ID_ANY, 909 SPDK_MALLOC_DMA); 910 if (!tqpair->bufs) { 911 SPDK_ERRLOG("Unable to allocate bufs on tqpair=%p.\n", tqpair); 912 return -1; 913 } 914 } 915 916 /* Add addtional one member, which will be used for mgmt_pdu owned by the tqpair */ 917 tqpair->pdus = spdk_dma_malloc((tqpair->resource_count + 1) * sizeof(*tqpair->pdus), 0x1000, NULL); 918 if (!tqpair->pdus) { 919 SPDK_ERRLOG("Unable to allocate pdu pool on tqpair =%p.\n", tqpair); 920 return -1; 921 } 922 923 for (i = 0; i < tqpair->resource_count; i++) { 924 struct spdk_nvmf_tcp_req *tcp_req = &tqpair->reqs[i]; 925 926 tcp_req->ttag = i + 1; 927 tcp_req->req.qpair = &tqpair->qpair; 928 929 tcp_req->pdu = &tqpair->pdus[i]; 930 tcp_req->pdu->qpair = tqpair; 931 932 /* Set up memory to receive commands */ 933 if (tqpair->bufs) { 934 tcp_req->buf = (void *)((uintptr_t)tqpair->bufs + (i * in_capsule_data_size)); 935 } 936 937 /* Set the cmdn and rsp */ 938 tcp_req->req.rsp = (union nvmf_c2h_msg *)&tcp_req->rsp; 939 tcp_req->req.cmd = (union nvmf_h2c_msg *)&tcp_req->cmd; 940 941 /* Initialize request state to FREE */ 942 tcp_req->state = TCP_REQUEST_STATE_FREE; 943 TAILQ_INSERT_TAIL(&tqpair->tcp_req_free_queue, tcp_req, state_link); 944 tqpair->state_cntr[TCP_REQUEST_STATE_FREE]++; 945 } 946 947 tqpair->mgmt_pdu = &tqpair->pdus[i]; 948 tqpair->mgmt_pdu->qpair = tqpair; 949 950 tqpair->recv_buf_size = (in_capsule_data_size + sizeof(struct spdk_nvme_tcp_cmd) + 2 * 951 SPDK_NVME_TCP_DIGEST_LEN) * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 952 953 return 0; 954 } 955 956 static int 957 nvmf_tcp_qpair_init(struct spdk_nvmf_qpair *qpair) 958 { 959 struct spdk_nvmf_tcp_qpair *tqpair; 960 961 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 962 963 SPDK_DEBUGLOG(nvmf_tcp, "New TCP Connection: %p\n", qpair); 964 965 /* Initialise request state queues of the qpair */ 966 TAILQ_INIT(&tqpair->tcp_req_free_queue); 967 TAILQ_INIT(&tqpair->tcp_req_working_queue); 968 969 tqpair->host_hdgst_enable = true; 970 tqpair->host_ddgst_enable = true; 971 972 return 0; 973 } 974 975 static int 976 nvmf_tcp_qpair_sock_init(struct spdk_nvmf_tcp_qpair *tqpair) 977 { 978 int rc; 979 980 /* set low water mark */ 981 rc = spdk_sock_set_recvlowat(tqpair->sock, sizeof(struct spdk_nvme_tcp_common_pdu_hdr)); 982 if (rc != 0) { 983 SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n"); 984 return rc; 985 } 986 987 return 0; 988 } 989 990 static void 991 nvmf_tcp_handle_connect(struct spdk_nvmf_transport *transport, 992 struct spdk_nvmf_tcp_port *port, 993 struct spdk_sock *sock) 994 { 995 struct spdk_nvmf_tcp_qpair *tqpair; 996 int rc; 997 998 SPDK_DEBUGLOG(nvmf_tcp, "New connection accepted on %s port %s\n", 999 port->trid->traddr, port->trid->trsvcid); 1000 1001 tqpair = calloc(1, sizeof(struct spdk_nvmf_tcp_qpair)); 1002 if (tqpair == NULL) { 1003 SPDK_ERRLOG("Could not allocate new connection.\n"); 1004 spdk_sock_close(&sock); 1005 return; 1006 } 1007 1008 tqpair->sock = sock; 1009 tqpair->state_cntr[TCP_REQUEST_STATE_FREE] = 0; 1010 tqpair->port = port; 1011 tqpair->qpair.transport = transport; 1012 1013 rc = spdk_sock_getaddr(tqpair->sock, tqpair->target_addr, 1014 sizeof(tqpair->target_addr), &tqpair->target_port, 1015 tqpair->initiator_addr, sizeof(tqpair->initiator_addr), 1016 &tqpair->initiator_port); 1017 if (rc < 0) { 1018 SPDK_ERRLOG("spdk_sock_getaddr() failed of tqpair=%p\n", tqpair); 1019 nvmf_tcp_qpair_destroy(tqpair); 1020 return; 1021 } 1022 1023 spdk_nvmf_tgt_new_qpair(transport->tgt, &tqpair->qpair); 1024 } 1025 1026 static uint32_t 1027 nvmf_tcp_port_accept(struct spdk_nvmf_transport *transport, struct spdk_nvmf_tcp_port *port) 1028 { 1029 struct spdk_sock *sock; 1030 uint32_t count = 0; 1031 int i; 1032 1033 for (i = 0; i < NVMF_TCP_MAX_ACCEPT_SOCK_ONE_TIME; i++) { 1034 sock = spdk_sock_accept(port->listen_sock); 1035 if (sock == NULL) { 1036 break; 1037 } 1038 count++; 1039 nvmf_tcp_handle_connect(transport, port, sock); 1040 } 1041 1042 return count; 1043 } 1044 1045 static uint32_t 1046 nvmf_tcp_accept(struct spdk_nvmf_transport *transport) 1047 { 1048 struct spdk_nvmf_tcp_transport *ttransport; 1049 struct spdk_nvmf_tcp_port *port; 1050 uint32_t count = 0; 1051 1052 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1053 1054 TAILQ_FOREACH(port, &ttransport->ports, link) { 1055 count += nvmf_tcp_port_accept(transport, port); 1056 } 1057 1058 return count; 1059 } 1060 1061 static void 1062 nvmf_tcp_discover(struct spdk_nvmf_transport *transport, 1063 struct spdk_nvme_transport_id *trid, 1064 struct spdk_nvmf_discovery_log_page_entry *entry) 1065 { 1066 entry->trtype = SPDK_NVMF_TRTYPE_TCP; 1067 entry->adrfam = trid->adrfam; 1068 entry->treq.secure_channel = SPDK_NVMF_TREQ_SECURE_CHANNEL_NOT_REQUIRED; 1069 1070 spdk_strcpy_pad(entry->trsvcid, trid->trsvcid, sizeof(entry->trsvcid), ' '); 1071 spdk_strcpy_pad(entry->traddr, trid->traddr, sizeof(entry->traddr), ' '); 1072 1073 entry->tsas.tcp.sectype = SPDK_NVME_TCP_SECURITY_NONE; 1074 } 1075 1076 static struct spdk_nvmf_tcp_control_msg_list * 1077 nvmf_tcp_control_msg_list_create(uint16_t num_messages) 1078 { 1079 struct spdk_nvmf_tcp_control_msg_list *list; 1080 struct spdk_nvmf_tcp_control_msg *msg; 1081 uint16_t i; 1082 1083 list = calloc(1, sizeof(*list)); 1084 if (!list) { 1085 SPDK_ERRLOG("Failed to allocate memory for list structure\n"); 1086 return NULL; 1087 } 1088 1089 list->msg_buf = spdk_zmalloc(num_messages * SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE, 1090 NVMF_DATA_BUFFER_ALIGNMENT, NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1091 if (!list->msg_buf) { 1092 SPDK_ERRLOG("Failed to allocate memory for control message buffers\n"); 1093 free(list); 1094 return NULL; 1095 } 1096 1097 STAILQ_INIT(&list->free_msgs); 1098 1099 for (i = 0; i < num_messages; i++) { 1100 msg = (struct spdk_nvmf_tcp_control_msg *)((char *)list->msg_buf + i * 1101 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1102 STAILQ_INSERT_TAIL(&list->free_msgs, msg, link); 1103 } 1104 1105 return list; 1106 } 1107 1108 static void 1109 nvmf_tcp_control_msg_list_free(struct spdk_nvmf_tcp_control_msg_list *list) 1110 { 1111 if (!list) { 1112 return; 1113 } 1114 1115 spdk_free(list->msg_buf); 1116 free(list); 1117 } 1118 1119 static struct spdk_nvmf_transport_poll_group * 1120 nvmf_tcp_poll_group_create(struct spdk_nvmf_transport *transport) 1121 { 1122 struct spdk_nvmf_tcp_transport *ttransport; 1123 struct spdk_nvmf_tcp_poll_group *tgroup; 1124 1125 tgroup = calloc(1, sizeof(*tgroup)); 1126 if (!tgroup) { 1127 return NULL; 1128 } 1129 1130 tgroup->sock_group = spdk_sock_group_create(&tgroup->group); 1131 if (!tgroup->sock_group) { 1132 goto cleanup; 1133 } 1134 1135 TAILQ_INIT(&tgroup->qpairs); 1136 TAILQ_INIT(&tgroup->await_req); 1137 1138 ttransport = SPDK_CONTAINEROF(transport, struct spdk_nvmf_tcp_transport, transport); 1139 1140 if (transport->opts.in_capsule_data_size < SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE) { 1141 SPDK_DEBUGLOG(nvmf_tcp, "ICD %u is less than min required for admin/fabric commands (%u). " 1142 "Creating control messages list\n", transport->opts.in_capsule_data_size, 1143 SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE); 1144 tgroup->control_msg_list = nvmf_tcp_control_msg_list_create(ttransport->tcp_opts.control_msg_num); 1145 if (!tgroup->control_msg_list) { 1146 goto cleanup; 1147 } 1148 } 1149 1150 return &tgroup->group; 1151 1152 cleanup: 1153 nvmf_tcp_poll_group_destroy(&tgroup->group); 1154 return NULL; 1155 } 1156 1157 static struct spdk_nvmf_transport_poll_group * 1158 nvmf_tcp_get_optimal_poll_group(struct spdk_nvmf_qpair *qpair) 1159 { 1160 struct spdk_nvmf_tcp_qpair *tqpair; 1161 struct spdk_sock_group *group = NULL; 1162 int rc; 1163 1164 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 1165 rc = spdk_sock_get_optimal_sock_group(tqpair->sock, &group); 1166 if (!rc && group != NULL) { 1167 return spdk_sock_group_get_ctx(group); 1168 } 1169 1170 return NULL; 1171 } 1172 1173 static void 1174 nvmf_tcp_poll_group_destroy(struct spdk_nvmf_transport_poll_group *group) 1175 { 1176 struct spdk_nvmf_tcp_poll_group *tgroup; 1177 1178 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 1179 spdk_sock_group_close(&tgroup->sock_group); 1180 if (tgroup->control_msg_list) { 1181 nvmf_tcp_control_msg_list_free(tgroup->control_msg_list); 1182 } 1183 1184 free(tgroup); 1185 } 1186 1187 static void 1188 nvmf_tcp_qpair_set_recv_state(struct spdk_nvmf_tcp_qpair *tqpair, 1189 enum nvme_tcp_pdu_recv_state state) 1190 { 1191 if (tqpair->recv_state == state) { 1192 SPDK_ERRLOG("The recv state of tqpair=%p is same with the state(%d) to be set\n", 1193 tqpair, state); 1194 return; 1195 } 1196 1197 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 1198 /* When leaving the await req state, move the qpair to the main list */ 1199 TAILQ_REMOVE(&tqpair->group->await_req, tqpair, link); 1200 TAILQ_INSERT_TAIL(&tqpair->group->qpairs, tqpair, link); 1201 } 1202 1203 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv state=%d\n", tqpair, state); 1204 tqpair->recv_state = state; 1205 1206 switch (state) { 1207 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1208 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1209 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1210 break; 1211 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1212 TAILQ_REMOVE(&tqpair->group->qpairs, tqpair, link); 1213 TAILQ_INSERT_TAIL(&tqpair->group->await_req, tqpair, link); 1214 break; 1215 case NVME_TCP_PDU_RECV_STATE_ERROR: 1216 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1217 memset(&tqpair->pdu_in_progress, 0, sizeof(tqpair->pdu_in_progress)); 1218 break; 1219 default: 1220 SPDK_ERRLOG("The state(%d) is invalid\n", state); 1221 abort(); 1222 break; 1223 } 1224 } 1225 1226 static int 1227 nvmf_tcp_qpair_handle_timeout(void *ctx) 1228 { 1229 struct spdk_nvmf_tcp_qpair *tqpair = ctx; 1230 1231 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_ERROR); 1232 1233 SPDK_ERRLOG("No pdu coming for tqpair=%p within %d seconds\n", tqpair, 1234 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT); 1235 1236 nvmf_tcp_qpair_disconnect(tqpair); 1237 return SPDK_POLLER_BUSY; 1238 } 1239 1240 static void 1241 nvmf_tcp_send_c2h_term_req_complete(void *cb_arg) 1242 { 1243 struct spdk_nvmf_tcp_qpair *tqpair = (struct spdk_nvmf_tcp_qpair *)cb_arg; 1244 1245 if (!tqpair->timeout_poller) { 1246 tqpair->timeout_poller = SPDK_POLLER_REGISTER(nvmf_tcp_qpair_handle_timeout, tqpair, 1247 SPDK_NVME_TCP_QPAIR_EXIT_TIMEOUT * 1000000); 1248 } 1249 } 1250 1251 static void 1252 nvmf_tcp_send_c2h_term_req(struct spdk_nvmf_tcp_qpair *tqpair, struct nvme_tcp_pdu *pdu, 1253 enum spdk_nvme_tcp_term_req_fes fes, uint32_t error_offset) 1254 { 1255 struct nvme_tcp_pdu *rsp_pdu; 1256 struct spdk_nvme_tcp_term_req_hdr *c2h_term_req; 1257 uint32_t c2h_term_req_hdr_len = sizeof(*c2h_term_req); 1258 uint32_t copy_len; 1259 1260 rsp_pdu = tqpair->mgmt_pdu; 1261 1262 c2h_term_req = &rsp_pdu->hdr.term_req; 1263 c2h_term_req->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_TERM_REQ; 1264 c2h_term_req->common.hlen = c2h_term_req_hdr_len; 1265 1266 if ((fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1267 (fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1268 DSET32(&c2h_term_req->fei, error_offset); 1269 } 1270 1271 copy_len = spdk_min(pdu->hdr.common.hlen, SPDK_NVME_TCP_TERM_REQ_ERROR_DATA_MAX_SIZE); 1272 1273 /* Copy the error info into the buffer */ 1274 memcpy((uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, pdu->hdr.raw, copy_len); 1275 nvme_tcp_pdu_set_data(rsp_pdu, (uint8_t *)rsp_pdu->hdr.raw + c2h_term_req_hdr_len, copy_len); 1276 1277 /* Contain the header of the wrong received pdu */ 1278 c2h_term_req->common.plen = c2h_term_req->common.hlen + copy_len; 1279 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1280 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_c2h_term_req_complete, tqpair); 1281 } 1282 1283 static void 1284 nvmf_tcp_capsule_cmd_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1285 struct spdk_nvmf_tcp_qpair *tqpair, 1286 struct nvme_tcp_pdu *pdu) 1287 { 1288 struct spdk_nvmf_tcp_req *tcp_req; 1289 1290 assert(pdu->psh_valid_bytes == pdu->psh_len); 1291 assert(pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD); 1292 1293 tcp_req = nvmf_tcp_req_get(tqpair); 1294 if (!tcp_req) { 1295 /* Directly return and make the allocation retry again */ 1296 if (tqpair->state_cntr[TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST] > 0) { 1297 return; 1298 } 1299 1300 /* The host sent more commands than the maximum queue depth. */ 1301 SPDK_ERRLOG("Cannot allocate tcp_req on tqpair=%p\n", tqpair); 1302 nvmf_tcp_qpair_disconnect(tqpair); 1303 return; 1304 } 1305 1306 pdu->req = tcp_req; 1307 assert(tcp_req->state == TCP_REQUEST_STATE_NEW); 1308 nvmf_tcp_req_process(ttransport, tcp_req); 1309 } 1310 1311 static void 1312 nvmf_tcp_capsule_cmd_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1313 struct spdk_nvmf_tcp_qpair *tqpair, 1314 struct nvme_tcp_pdu *pdu) 1315 { 1316 struct spdk_nvmf_tcp_req *tcp_req; 1317 struct spdk_nvme_tcp_cmd *capsule_cmd; 1318 uint32_t error_offset = 0; 1319 enum spdk_nvme_tcp_term_req_fes fes; 1320 1321 capsule_cmd = &pdu->hdr.capsule_cmd; 1322 tcp_req = pdu->req; 1323 assert(tcp_req != NULL); 1324 if (capsule_cmd->common.pdo > SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET) { 1325 SPDK_ERRLOG("Expected ICReq capsule_cmd pdu offset <= %d, got %c\n", 1326 SPDK_NVME_TCP_PDU_PDO_MAX_OFFSET, capsule_cmd->common.pdo); 1327 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1328 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1329 goto err; 1330 } 1331 1332 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1333 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1334 nvmf_tcp_req_process(ttransport, tcp_req); 1335 1336 return; 1337 err: 1338 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1339 } 1340 1341 static int 1342 nvmf_tcp_find_req_in_state(struct spdk_nvmf_tcp_qpair *tqpair, 1343 enum spdk_nvmf_tcp_req_state state, 1344 uint16_t cid, uint16_t tag, 1345 struct spdk_nvmf_tcp_req **req) 1346 { 1347 struct spdk_nvmf_tcp_req *tcp_req = NULL; 1348 1349 TAILQ_FOREACH(tcp_req, &tqpair->tcp_req_working_queue, state_link) { 1350 if (tcp_req->state != state) { 1351 continue; 1352 } 1353 1354 if (tcp_req->req.cmd->nvme_cmd.cid != cid) { 1355 continue; 1356 } 1357 1358 if (tcp_req->ttag == tag) { 1359 *req = tcp_req; 1360 return 0; 1361 } 1362 1363 *req = NULL; 1364 return -1; 1365 } 1366 1367 /* Didn't find it, but not an error */ 1368 *req = NULL; 1369 return 0; 1370 } 1371 1372 static void 1373 nvmf_tcp_h2c_data_hdr_handle(struct spdk_nvmf_tcp_transport *ttransport, 1374 struct spdk_nvmf_tcp_qpair *tqpair, 1375 struct nvme_tcp_pdu *pdu) 1376 { 1377 struct spdk_nvmf_tcp_req *tcp_req; 1378 uint32_t error_offset = 0; 1379 enum spdk_nvme_tcp_term_req_fes fes = 0; 1380 struct spdk_nvme_tcp_h2c_data_hdr *h2c_data; 1381 int rc; 1382 1383 h2c_data = &pdu->hdr.h2c_data; 1384 1385 SPDK_DEBUGLOG(nvmf_tcp, "tqpair=%p, r2t_info: datao=%u, datal=%u, cccid=%u, ttag=%u\n", 1386 tqpair, h2c_data->datao, h2c_data->datal, h2c_data->cccid, h2c_data->ttag); 1387 1388 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 1389 h2c_data->cccid, h2c_data->ttag, &tcp_req); 1390 if (rc == 0 && tcp_req == NULL) { 1391 rc = nvmf_tcp_find_req_in_state(tqpair, TCP_REQUEST_STATE_AWAITING_R2T_ACK, h2c_data->cccid, 1392 h2c_data->ttag, &tcp_req); 1393 } 1394 1395 if (!tcp_req) { 1396 SPDK_DEBUGLOG(nvmf_tcp, "tcp_req is not found for tqpair=%p\n", tqpair); 1397 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER; 1398 if (rc == 0) { 1399 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, cccid); 1400 } else { 1401 error_offset = offsetof(struct spdk_nvme_tcp_h2c_data_hdr, ttag); 1402 } 1403 goto err; 1404 } 1405 1406 if (tcp_req->h2c_offset != h2c_data->datao) { 1407 SPDK_DEBUGLOG(nvmf_tcp, 1408 "tcp_req(%p), tqpair=%p, expected data offset %u, but data offset is %u\n", 1409 tcp_req, tqpair, tcp_req->h2c_offset, h2c_data->datao); 1410 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1411 goto err; 1412 } 1413 1414 if ((h2c_data->datao + h2c_data->datal) > tcp_req->req.length) { 1415 SPDK_DEBUGLOG(nvmf_tcp, 1416 "tcp_req(%p), tqpair=%p, (datao=%u + datal=%u) execeeds requested length=%u\n", 1417 tcp_req, tqpair, h2c_data->datao, h2c_data->datal, tcp_req->req.length); 1418 fes = SPDK_NVME_TCP_TERM_REQ_FES_DATA_TRANSFER_OUT_OF_RANGE; 1419 goto err; 1420 } 1421 1422 pdu->req = tcp_req; 1423 1424 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 1425 pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 1426 } 1427 1428 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 1429 h2c_data->datao, h2c_data->datal); 1430 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1431 return; 1432 1433 err: 1434 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1435 } 1436 1437 static void 1438 nvmf_tcp_send_capsule_resp_pdu(struct spdk_nvmf_tcp_req *tcp_req, 1439 struct spdk_nvmf_tcp_qpair *tqpair) 1440 { 1441 struct nvme_tcp_pdu *rsp_pdu; 1442 struct spdk_nvme_tcp_rsp *capsule_resp; 1443 1444 SPDK_DEBUGLOG(nvmf_tcp, "enter, tqpair=%p\n", tqpair); 1445 1446 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1447 assert(rsp_pdu != NULL); 1448 1449 capsule_resp = &rsp_pdu->hdr.capsule_resp; 1450 capsule_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_CAPSULE_RESP; 1451 capsule_resp->common.plen = capsule_resp->common.hlen = sizeof(*capsule_resp); 1452 capsule_resp->rccqe = tcp_req->req.rsp->nvme_cpl; 1453 if (tqpair->host_hdgst_enable) { 1454 capsule_resp->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1455 capsule_resp->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1456 } 1457 1458 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_request_free, tcp_req); 1459 } 1460 1461 static void 1462 nvmf_tcp_pdu_c2h_data_complete(void *cb_arg) 1463 { 1464 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1465 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, 1466 struct spdk_nvmf_tcp_qpair, qpair); 1467 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 1468 tcp_req->req.qpair->transport, struct spdk_nvmf_tcp_transport, transport); 1469 1470 assert(tqpair != NULL); 1471 if (ttransport->tcp_opts.c2h_success) { 1472 nvmf_tcp_request_free(tcp_req); 1473 } else { 1474 nvmf_tcp_req_pdu_fini(tcp_req); 1475 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 1476 } 1477 } 1478 1479 static void 1480 nvmf_tcp_r2t_complete(void *cb_arg) 1481 { 1482 struct spdk_nvmf_tcp_req *tcp_req = cb_arg; 1483 struct spdk_nvmf_tcp_transport *ttransport; 1484 1485 nvmf_tcp_req_pdu_fini(tcp_req); 1486 1487 ttransport = SPDK_CONTAINEROF(tcp_req->req.qpair->transport, 1488 struct spdk_nvmf_tcp_transport, transport); 1489 1490 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 1491 1492 if (tcp_req->h2c_offset == tcp_req->req.length) { 1493 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1494 nvmf_tcp_req_process(ttransport, tcp_req); 1495 } 1496 } 1497 1498 static void 1499 nvmf_tcp_send_r2t_pdu(struct spdk_nvmf_tcp_qpair *tqpair, 1500 struct spdk_nvmf_tcp_req *tcp_req) 1501 { 1502 struct nvme_tcp_pdu *rsp_pdu; 1503 struct spdk_nvme_tcp_r2t_hdr *r2t; 1504 1505 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 1506 assert(rsp_pdu != NULL); 1507 1508 r2t = &rsp_pdu->hdr.r2t; 1509 r2t->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_R2T; 1510 r2t->common.plen = r2t->common.hlen = sizeof(*r2t); 1511 1512 if (tqpair->host_hdgst_enable) { 1513 r2t->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 1514 r2t->common.plen += SPDK_NVME_TCP_DIGEST_LEN; 1515 } 1516 1517 r2t->cccid = tcp_req->req.cmd->nvme_cmd.cid; 1518 r2t->ttag = tcp_req->ttag; 1519 r2t->r2to = tcp_req->h2c_offset; 1520 r2t->r2tl = tcp_req->req.length; 1521 1522 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_AWAITING_R2T_ACK); 1523 1524 SPDK_DEBUGLOG(nvmf_tcp, 1525 "tcp_req(%p) on tqpair(%p), r2t_info: cccid=%u, ttag=%u, r2to=%u, r2tl=%u\n", 1526 tcp_req, tqpair, r2t->cccid, r2t->ttag, r2t->r2to, r2t->r2tl); 1527 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_r2t_complete, tcp_req); 1528 } 1529 1530 static void 1531 nvmf_tcp_h2c_data_payload_handle(struct spdk_nvmf_tcp_transport *ttransport, 1532 struct spdk_nvmf_tcp_qpair *tqpair, 1533 struct nvme_tcp_pdu *pdu) 1534 { 1535 struct spdk_nvmf_tcp_req *tcp_req; 1536 1537 tcp_req = pdu->req; 1538 assert(tcp_req != NULL); 1539 1540 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1541 1542 tcp_req->h2c_offset += pdu->data_len; 1543 1544 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1545 1546 /* Wait for all of the data to arrive AND for the initial R2T PDU send to be 1547 * acknowledged before moving on. */ 1548 if (tcp_req->h2c_offset == tcp_req->req.length && 1549 tcp_req->state == TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER) { 1550 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 1551 nvmf_tcp_req_process(ttransport, tcp_req); 1552 } 1553 } 1554 1555 static void 1556 nvmf_tcp_h2c_term_req_dump(struct spdk_nvme_tcp_term_req_hdr *h2c_term_req) 1557 { 1558 SPDK_ERRLOG("Error info of pdu(%p): %s\n", h2c_term_req, 1559 spdk_nvmf_tcp_term_req_fes_str[h2c_term_req->fes]); 1560 if ((h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD) || 1561 (h2c_term_req->fes == SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER)) { 1562 SPDK_DEBUGLOG(nvmf_tcp, "The offset from the start of the PDU header is %u\n", 1563 DGET32(h2c_term_req->fei)); 1564 } 1565 } 1566 1567 static void 1568 nvmf_tcp_h2c_term_req_hdr_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1569 struct nvme_tcp_pdu *pdu) 1570 { 1571 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1572 uint32_t error_offset = 0; 1573 enum spdk_nvme_tcp_term_req_fes fes; 1574 1575 if (h2c_term_req->fes > SPDK_NVME_TCP_TERM_REQ_FES_INVALID_DATA_UNSUPPORTED_PARAMETER) { 1576 SPDK_ERRLOG("Fatal Error Status(FES) is unknown for h2c_term_req pdu=%p\n", pdu); 1577 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1578 error_offset = offsetof(struct spdk_nvme_tcp_term_req_hdr, fes); 1579 goto end; 1580 } 1581 1582 /* set the data buffer */ 1583 nvme_tcp_pdu_set_data(pdu, (uint8_t *)pdu->hdr.raw + h2c_term_req->common.hlen, 1584 h2c_term_req->common.plen - h2c_term_req->common.hlen); 1585 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1586 return; 1587 end: 1588 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1589 } 1590 1591 static void 1592 nvmf_tcp_h2c_term_req_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1593 struct nvme_tcp_pdu *pdu) 1594 { 1595 struct spdk_nvme_tcp_term_req_hdr *h2c_term_req = &pdu->hdr.term_req; 1596 1597 nvmf_tcp_h2c_term_req_dump(h2c_term_req); 1598 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 1599 } 1600 1601 static void 1602 nvmf_tcp_pdu_payload_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1603 struct spdk_nvmf_tcp_transport *ttransport) 1604 { 1605 int rc = 0; 1606 struct nvme_tcp_pdu *pdu; 1607 uint32_t crc32c, error_offset = 0; 1608 enum spdk_nvme_tcp_term_req_fes fes; 1609 1610 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 1611 pdu = &tqpair->pdu_in_progress; 1612 1613 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 1614 /* check data digest if need */ 1615 if (pdu->ddgst_enable) { 1616 crc32c = nvme_tcp_pdu_calc_data_digest(pdu); 1617 rc = MATCH_DIGEST_WORD(pdu->data_digest, crc32c); 1618 if (rc == 0) { 1619 SPDK_ERRLOG("Data digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1620 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1621 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1622 return; 1623 1624 } 1625 } 1626 1627 switch (pdu->hdr.common.pdu_type) { 1628 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1629 nvmf_tcp_capsule_cmd_payload_handle(ttransport, tqpair, pdu); 1630 break; 1631 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1632 nvmf_tcp_h2c_data_payload_handle(ttransport, tqpair, pdu); 1633 break; 1634 1635 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1636 nvmf_tcp_h2c_term_req_payload_handle(tqpair, pdu); 1637 break; 1638 1639 default: 1640 /* The code should not go to here */ 1641 SPDK_ERRLOG("The code should not go to here\n"); 1642 break; 1643 } 1644 } 1645 1646 static void 1647 nvmf_tcp_send_icresp_complete(void *cb_arg) 1648 { 1649 struct spdk_nvmf_tcp_qpair *tqpair = cb_arg; 1650 1651 tqpair->state = NVME_TCP_QPAIR_STATE_RUNNING; 1652 } 1653 1654 static void 1655 nvmf_tcp_icreq_handle(struct spdk_nvmf_tcp_transport *ttransport, 1656 struct spdk_nvmf_tcp_qpair *tqpair, 1657 struct nvme_tcp_pdu *pdu) 1658 { 1659 struct spdk_nvme_tcp_ic_req *ic_req = &pdu->hdr.ic_req; 1660 struct nvme_tcp_pdu *rsp_pdu; 1661 struct spdk_nvme_tcp_ic_resp *ic_resp; 1662 uint32_t error_offset = 0; 1663 enum spdk_nvme_tcp_term_req_fes fes; 1664 1665 /* Only PFV 0 is defined currently */ 1666 if (ic_req->pfv != 0) { 1667 SPDK_ERRLOG("Expected ICReq PFV %u, got %u\n", 0u, ic_req->pfv); 1668 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1669 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, pfv); 1670 goto end; 1671 } 1672 1673 /* MAXR2T is 0's based */ 1674 SPDK_DEBUGLOG(nvmf_tcp, "maxr2t =%u\n", (ic_req->maxr2t + 1u)); 1675 1676 tqpair->host_hdgst_enable = ic_req->dgst.bits.hdgst_enable ? true : false; 1677 if (!tqpair->host_hdgst_enable) { 1678 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1679 } 1680 1681 tqpair->host_ddgst_enable = ic_req->dgst.bits.ddgst_enable ? true : false; 1682 if (!tqpair->host_ddgst_enable) { 1683 tqpair->recv_buf_size -= SPDK_NVME_TCP_DIGEST_LEN * SPDK_NVMF_TCP_RECV_BUF_SIZE_FACTOR; 1684 } 1685 1686 tqpair->recv_buf_size = spdk_max(tqpair->recv_buf_size, MIN_SOCK_PIPE_SIZE); 1687 /* Now that we know whether digests are enabled, properly size the receive buffer */ 1688 if (spdk_sock_set_recvbuf(tqpair->sock, tqpair->recv_buf_size) < 0) { 1689 SPDK_WARNLOG("Unable to allocate enough memory for receive buffer on tqpair=%p with size=%d\n", 1690 tqpair, 1691 tqpair->recv_buf_size); 1692 /* Not fatal. */ 1693 } 1694 1695 if (tqpair->host_hdgst_enable) { 1696 tqpair->accel_channel = spdk_accel_engine_get_io_channel(); 1697 if (spdk_unlikely(!tqpair->accel_channel)) { 1698 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1699 error_offset = offsetof(struct spdk_nvme_tcp_ic_req, dgst); 1700 SPDK_ERRLOG("Unabled to get accel_channel for tqpair=%p, failed to enable header digest\n", 1701 tqpair); 1702 goto end; 1703 } 1704 } 1705 1706 tqpair->cpda = spdk_min(ic_req->hpda, SPDK_NVME_TCP_CPDA_MAX); 1707 SPDK_DEBUGLOG(nvmf_tcp, "cpda of tqpair=(%p) is : %u\n", tqpair, tqpair->cpda); 1708 1709 rsp_pdu = tqpair->mgmt_pdu; 1710 1711 ic_resp = &rsp_pdu->hdr.ic_resp; 1712 ic_resp->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_IC_RESP; 1713 ic_resp->common.hlen = ic_resp->common.plen = sizeof(*ic_resp); 1714 ic_resp->pfv = 0; 1715 ic_resp->cpda = tqpair->cpda; 1716 ic_resp->maxh2cdata = ttransport->transport.opts.max_io_size; 1717 ic_resp->dgst.bits.hdgst_enable = tqpair->host_hdgst_enable ? 1 : 0; 1718 ic_resp->dgst.bits.ddgst_enable = tqpair->host_ddgst_enable ? 1 : 0; 1719 1720 SPDK_DEBUGLOG(nvmf_tcp, "host_hdgst_enable: %u\n", tqpair->host_hdgst_enable); 1721 SPDK_DEBUGLOG(nvmf_tcp, "host_ddgst_enable: %u\n", tqpair->host_ddgst_enable); 1722 1723 tqpair->state = NVME_TCP_QPAIR_STATE_INITIALIZING; 1724 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_send_icresp_complete, tqpair); 1725 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 1726 return; 1727 end: 1728 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1729 } 1730 1731 static void 1732 nvmf_tcp_pdu_psh_handle(struct spdk_nvmf_tcp_qpair *tqpair, 1733 struct spdk_nvmf_tcp_transport *ttransport) 1734 { 1735 struct nvme_tcp_pdu *pdu; 1736 int rc; 1737 uint32_t crc32c, error_offset = 0; 1738 enum spdk_nvme_tcp_term_req_fes fes; 1739 1740 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1741 pdu = &tqpair->pdu_in_progress; 1742 1743 SPDK_DEBUGLOG(nvmf_tcp, "pdu type of tqpair(%p) is %d\n", tqpair, 1744 pdu->hdr.common.pdu_type); 1745 /* check header digest if needed */ 1746 if (pdu->has_hdgst) { 1747 SPDK_DEBUGLOG(nvmf_tcp, "Compare the header of pdu=%p on tqpair=%p\n", pdu, tqpair); 1748 crc32c = nvme_tcp_pdu_calc_header_digest(pdu); 1749 rc = MATCH_DIGEST_WORD((uint8_t *)pdu->hdr.raw + pdu->hdr.common.hlen, crc32c); 1750 if (rc == 0) { 1751 SPDK_ERRLOG("Header digest error on tqpair=(%p) with pdu=%p\n", tqpair, pdu); 1752 fes = SPDK_NVME_TCP_TERM_REQ_FES_HDGST_ERROR; 1753 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1754 return; 1755 1756 } 1757 } 1758 1759 switch (pdu->hdr.common.pdu_type) { 1760 case SPDK_NVME_TCP_PDU_TYPE_IC_REQ: 1761 nvmf_tcp_icreq_handle(ttransport, tqpair, pdu); 1762 break; 1763 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1764 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_REQ); 1765 break; 1766 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1767 nvmf_tcp_h2c_data_hdr_handle(ttransport, tqpair, pdu); 1768 break; 1769 1770 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1771 nvmf_tcp_h2c_term_req_hdr_handle(tqpair, pdu); 1772 break; 1773 1774 default: 1775 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", tqpair->pdu_in_progress.hdr.common.pdu_type); 1776 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1777 error_offset = 1; 1778 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1779 break; 1780 } 1781 } 1782 1783 static void 1784 nvmf_tcp_pdu_ch_handle(struct spdk_nvmf_tcp_qpair *tqpair) 1785 { 1786 struct nvme_tcp_pdu *pdu; 1787 uint32_t error_offset = 0; 1788 enum spdk_nvme_tcp_term_req_fes fes; 1789 uint8_t expected_hlen, pdo; 1790 bool plen_error = false, pdo_error = false; 1791 1792 assert(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1793 pdu = &tqpair->pdu_in_progress; 1794 1795 if (pdu->hdr.common.pdu_type == SPDK_NVME_TCP_PDU_TYPE_IC_REQ) { 1796 if (tqpair->state != NVME_TCP_QPAIR_STATE_INVALID) { 1797 SPDK_ERRLOG("Already received ICreq PDU, and reject this pdu=%p\n", pdu); 1798 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1799 goto err; 1800 } 1801 expected_hlen = sizeof(struct spdk_nvme_tcp_ic_req); 1802 if (pdu->hdr.common.plen != expected_hlen) { 1803 plen_error = true; 1804 } 1805 } else { 1806 if (tqpair->state != NVME_TCP_QPAIR_STATE_RUNNING) { 1807 SPDK_ERRLOG("The TCP/IP connection is not negotitated\n"); 1808 fes = SPDK_NVME_TCP_TERM_REQ_FES_PDU_SEQUENCE_ERROR; 1809 goto err; 1810 } 1811 1812 switch (pdu->hdr.common.pdu_type) { 1813 case SPDK_NVME_TCP_PDU_TYPE_CAPSULE_CMD: 1814 expected_hlen = sizeof(struct spdk_nvme_tcp_cmd); 1815 pdo = pdu->hdr.common.pdo; 1816 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1817 pdo_error = true; 1818 break; 1819 } 1820 1821 if (pdu->hdr.common.plen < expected_hlen) { 1822 plen_error = true; 1823 } 1824 break; 1825 case SPDK_NVME_TCP_PDU_TYPE_H2C_DATA: 1826 expected_hlen = sizeof(struct spdk_nvme_tcp_h2c_data_hdr); 1827 pdo = pdu->hdr.common.pdo; 1828 if ((tqpair->cpda != 0) && (pdo != ((tqpair->cpda + 1) << 2))) { 1829 pdo_error = true; 1830 break; 1831 } 1832 if (pdu->hdr.common.plen < expected_hlen) { 1833 plen_error = true; 1834 } 1835 break; 1836 1837 case SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ: 1838 expected_hlen = sizeof(struct spdk_nvme_tcp_term_req_hdr); 1839 if ((pdu->hdr.common.plen <= expected_hlen) || 1840 (pdu->hdr.common.plen > SPDK_NVME_TCP_TERM_REQ_PDU_MAX_SIZE)) { 1841 plen_error = true; 1842 } 1843 break; 1844 1845 default: 1846 SPDK_ERRLOG("Unexpected PDU type 0x%02x\n", pdu->hdr.common.pdu_type); 1847 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1848 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdu_type); 1849 goto err; 1850 } 1851 } 1852 1853 if (pdu->hdr.common.hlen != expected_hlen) { 1854 SPDK_ERRLOG("PDU type=0x%02x, Expected ICReq header length %u, got %u on tqpair=%p\n", 1855 pdu->hdr.common.pdu_type, 1856 expected_hlen, pdu->hdr.common.hlen, tqpair); 1857 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1858 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, hlen); 1859 goto err; 1860 } else if (pdo_error) { 1861 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1862 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, pdo); 1863 } else if (plen_error) { 1864 fes = SPDK_NVME_TCP_TERM_REQ_FES_INVALID_HEADER_FIELD; 1865 error_offset = offsetof(struct spdk_nvme_tcp_common_pdu_hdr, plen); 1866 goto err; 1867 } else { 1868 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH); 1869 nvme_tcp_pdu_calc_psh_len(&tqpair->pdu_in_progress, tqpair->host_hdgst_enable); 1870 return; 1871 } 1872 err: 1873 nvmf_tcp_send_c2h_term_req(tqpair, pdu, fes, error_offset); 1874 } 1875 1876 static int 1877 nvmf_tcp_pdu_payload_insert_dif(struct nvme_tcp_pdu *pdu, uint32_t read_offset, 1878 int read_len) 1879 { 1880 int rc; 1881 1882 rc = spdk_dif_generate_stream(pdu->data_iov, pdu->data_iovcnt, 1883 read_offset, read_len, pdu->dif_ctx); 1884 if (rc != 0) { 1885 SPDK_ERRLOG("DIF generate failed\n"); 1886 } 1887 1888 return rc; 1889 } 1890 1891 static int 1892 nvmf_tcp_sock_process(struct spdk_nvmf_tcp_qpair *tqpair) 1893 { 1894 int rc = 0; 1895 struct nvme_tcp_pdu *pdu; 1896 enum nvme_tcp_pdu_recv_state prev_state; 1897 uint32_t data_len; 1898 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(tqpair->qpair.transport, 1899 struct spdk_nvmf_tcp_transport, transport); 1900 1901 /* The loop here is to allow for several back-to-back state changes. */ 1902 do { 1903 prev_state = tqpair->recv_state; 1904 SPDK_DEBUGLOG(nvmf_tcp, "tqpair(%p) recv pdu entering state %d\n", tqpair, prev_state); 1905 1906 pdu = &tqpair->pdu_in_progress; 1907 switch (tqpair->recv_state) { 1908 /* Wait for the common header */ 1909 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY: 1910 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH: 1911 if (spdk_unlikely(tqpair->state == NVME_TCP_QPAIR_STATE_INITIALIZING)) { 1912 return rc; 1913 } 1914 1915 rc = nvme_tcp_read_data(tqpair->sock, 1916 sizeof(struct spdk_nvme_tcp_common_pdu_hdr) - pdu->ch_valid_bytes, 1917 (void *)&pdu->hdr.common + pdu->ch_valid_bytes); 1918 if (rc < 0) { 1919 SPDK_DEBUGLOG(nvmf_tcp, "will disconnect tqpair=%p\n", tqpair); 1920 return NVME_TCP_PDU_FATAL; 1921 } else if (rc > 0) { 1922 pdu->ch_valid_bytes += rc; 1923 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 0, rc, 0, 0); 1924 if (spdk_likely(tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY)) { 1925 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_CH); 1926 } 1927 } 1928 1929 if (pdu->ch_valid_bytes < sizeof(struct spdk_nvme_tcp_common_pdu_hdr)) { 1930 return NVME_TCP_PDU_IN_PROGRESS; 1931 } 1932 1933 /* The command header of this PDU has now been read from the socket. */ 1934 nvmf_tcp_pdu_ch_handle(tqpair); 1935 break; 1936 /* Wait for the pdu specific header */ 1937 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PSH: 1938 rc = nvme_tcp_read_data(tqpair->sock, 1939 pdu->psh_len - pdu->psh_valid_bytes, 1940 (void *)&pdu->hdr.raw + sizeof(struct spdk_nvme_tcp_common_pdu_hdr) + pdu->psh_valid_bytes); 1941 if (rc < 0) { 1942 return NVME_TCP_PDU_FATAL; 1943 } else if (rc > 0) { 1944 spdk_trace_record(TRACE_TCP_READ_FROM_SOCKET_DONE, 1945 0, rc, 0, 0); 1946 pdu->psh_valid_bytes += rc; 1947 } 1948 1949 if (pdu->psh_valid_bytes < pdu->psh_len) { 1950 return NVME_TCP_PDU_IN_PROGRESS; 1951 } 1952 1953 /* All header(ch, psh, head digist) of this PDU has now been read from the socket. */ 1954 nvmf_tcp_pdu_psh_handle(tqpair, ttransport); 1955 break; 1956 /* Wait for the req slot */ 1957 case NVME_TCP_PDU_RECV_STATE_AWAIT_REQ: 1958 nvmf_tcp_capsule_cmd_hdr_handle(ttransport, tqpair, pdu); 1959 break; 1960 case NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD: 1961 /* check whether the data is valid, if not we just return */ 1962 if (!pdu->data_len) { 1963 return NVME_TCP_PDU_IN_PROGRESS; 1964 } 1965 1966 data_len = pdu->data_len; 1967 /* data digest */ 1968 if (spdk_unlikely((pdu->hdr.common.pdu_type != SPDK_NVME_TCP_PDU_TYPE_H2C_TERM_REQ) && 1969 tqpair->host_ddgst_enable)) { 1970 data_len += SPDK_NVME_TCP_DIGEST_LEN; 1971 pdu->ddgst_enable = true; 1972 } 1973 1974 rc = nvme_tcp_read_payload_data(tqpair->sock, pdu); 1975 if (rc < 0) { 1976 return NVME_TCP_PDU_FATAL; 1977 } 1978 pdu->readv_offset += rc; 1979 1980 if (spdk_unlikely(pdu->dif_ctx != NULL)) { 1981 rc = nvmf_tcp_pdu_payload_insert_dif(pdu, pdu->readv_offset - rc, rc); 1982 if (rc != 0) { 1983 return NVME_TCP_PDU_FATAL; 1984 } 1985 } 1986 1987 if (pdu->readv_offset < data_len) { 1988 return NVME_TCP_PDU_IN_PROGRESS; 1989 } 1990 1991 /* All of this PDU has now been read from the socket. */ 1992 nvmf_tcp_pdu_payload_handle(tqpair, ttransport); 1993 break; 1994 case NVME_TCP_PDU_RECV_STATE_ERROR: 1995 if (!spdk_sock_is_connected(tqpair->sock)) { 1996 return NVME_TCP_PDU_FATAL; 1997 } 1998 break; 1999 default: 2000 assert(0); 2001 SPDK_ERRLOG("code should not come to here"); 2002 break; 2003 } 2004 } while (tqpair->recv_state != prev_state); 2005 2006 return rc; 2007 } 2008 2009 static inline void * 2010 nvmf_tcp_control_msg_get(struct spdk_nvmf_tcp_control_msg_list *list) 2011 { 2012 struct spdk_nvmf_tcp_control_msg *msg; 2013 2014 assert(list); 2015 2016 msg = STAILQ_FIRST(&list->free_msgs); 2017 if (!msg) { 2018 SPDK_DEBUGLOG(nvmf_tcp, "Out of control messages\n"); 2019 return NULL; 2020 } 2021 STAILQ_REMOVE_HEAD(&list->free_msgs, link); 2022 return msg; 2023 } 2024 2025 static inline void 2026 nvmf_tcp_control_msg_put(struct spdk_nvmf_tcp_control_msg_list *list, void *_msg) 2027 { 2028 struct spdk_nvmf_tcp_control_msg *msg = _msg; 2029 2030 assert(list); 2031 STAILQ_INSERT_HEAD(&list->free_msgs, msg, link); 2032 } 2033 2034 static int 2035 nvmf_tcp_req_parse_sgl(struct spdk_nvmf_tcp_req *tcp_req, 2036 struct spdk_nvmf_transport *transport, 2037 struct spdk_nvmf_transport_poll_group *group) 2038 { 2039 struct spdk_nvmf_request *req = &tcp_req->req; 2040 struct spdk_nvme_cmd *cmd; 2041 struct spdk_nvme_cpl *rsp; 2042 struct spdk_nvme_sgl_descriptor *sgl; 2043 struct spdk_nvmf_tcp_poll_group *tgroup; 2044 uint32_t length; 2045 2046 cmd = &req->cmd->nvme_cmd; 2047 rsp = &req->rsp->nvme_cpl; 2048 sgl = &cmd->dptr.sgl1; 2049 2050 length = sgl->unkeyed.length; 2051 2052 if (sgl->generic.type == SPDK_NVME_SGL_TYPE_TRANSPORT_DATA_BLOCK && 2053 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_TRANSPORT) { 2054 if (length > transport->opts.max_io_size) { 2055 SPDK_ERRLOG("SGL length 0x%x exceeds max io size 0x%x\n", 2056 length, transport->opts.max_io_size); 2057 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2058 return -1; 2059 } 2060 2061 /* fill request length and populate iovs */ 2062 req->length = length; 2063 2064 SPDK_DEBUGLOG(nvmf_tcp, "Data requested length= 0x%x\n", length); 2065 2066 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2067 req->dif.orig_length = length; 2068 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2069 req->dif.elba_length = length; 2070 } 2071 2072 if (spdk_nvmf_request_get_buffers(req, group, transport, length)) { 2073 /* No available buffers. Queue this request up. */ 2074 SPDK_DEBUGLOG(nvmf_tcp, "No available large data buffers. Queueing request %p\n", 2075 tcp_req); 2076 return 0; 2077 } 2078 2079 /* backward compatible */ 2080 req->data = req->iov[0].iov_base; 2081 2082 SPDK_DEBUGLOG(nvmf_tcp, "Request %p took %d buffer/s from central pool, and data=%p\n", 2083 tcp_req, req->iovcnt, req->data); 2084 2085 return 0; 2086 } else if (sgl->generic.type == SPDK_NVME_SGL_TYPE_DATA_BLOCK && 2087 sgl->unkeyed.subtype == SPDK_NVME_SGL_SUBTYPE_OFFSET) { 2088 uint64_t offset = sgl->address; 2089 uint32_t max_len = transport->opts.in_capsule_data_size; 2090 assert(tcp_req->has_incapsule_data); 2091 2092 SPDK_DEBUGLOG(nvmf_tcp, "In-capsule data: offset 0x%" PRIx64 ", length 0x%x\n", 2093 offset, length); 2094 2095 if (offset > max_len) { 2096 SPDK_ERRLOG("In-capsule offset 0x%" PRIx64 " exceeds capsule length 0x%x\n", 2097 offset, max_len); 2098 rsp->status.sc = SPDK_NVME_SC_INVALID_SGL_OFFSET; 2099 return -1; 2100 } 2101 max_len -= (uint32_t)offset; 2102 2103 if (spdk_unlikely(length > max_len)) { 2104 /* According to the SPEC we should support ICD up to 8192 bytes for admin and fabric commands */ 2105 if (length <= SPDK_NVME_TCP_IN_CAPSULE_DATA_MAX_SIZE && 2106 (cmd->opc == SPDK_NVME_OPC_FABRIC || req->qpair->qid == 0)) { 2107 2108 /* Get a buffer from dedicated list */ 2109 SPDK_DEBUGLOG(nvmf_tcp, "Getting a buffer from control msg list\n"); 2110 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2111 assert(tgroup->control_msg_list); 2112 req->data = nvmf_tcp_control_msg_get(tgroup->control_msg_list); 2113 if (!req->data) { 2114 /* No available buffers. Queue this request up. */ 2115 SPDK_DEBUGLOG(nvmf_tcp, "No available ICD buffers. Queueing request %p\n", tcp_req); 2116 return 0; 2117 } 2118 } else { 2119 SPDK_ERRLOG("In-capsule data length 0x%x exceeds capsule length 0x%x\n", 2120 length, max_len); 2121 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 2122 return -1; 2123 } 2124 } else { 2125 req->data = tcp_req->buf; 2126 } 2127 2128 req->length = length; 2129 req->data_from_pool = false; 2130 2131 if (spdk_unlikely(req->dif.dif_insert_or_strip)) { 2132 length = spdk_dif_get_length_with_md(length, &req->dif.dif_ctx); 2133 req->dif.elba_length = length; 2134 } 2135 2136 req->iov[0].iov_base = req->data; 2137 req->iov[0].iov_len = length; 2138 req->iovcnt = 1; 2139 2140 return 0; 2141 } 2142 2143 SPDK_ERRLOG("Invalid NVMf I/O Command SGL: Type 0x%x, Subtype 0x%x\n", 2144 sgl->generic.type, sgl->generic.subtype); 2145 rsp->status.sc = SPDK_NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID; 2146 return -1; 2147 } 2148 2149 static inline enum spdk_nvme_media_error_status_code 2150 nvmf_tcp_dif_error_to_compl_status(uint8_t err_type) { 2151 enum spdk_nvme_media_error_status_code result; 2152 2153 switch (err_type) 2154 { 2155 case SPDK_DIF_REFTAG_ERROR: 2156 result = SPDK_NVME_SC_REFERENCE_TAG_CHECK_ERROR; 2157 break; 2158 case SPDK_DIF_APPTAG_ERROR: 2159 result = SPDK_NVME_SC_APPLICATION_TAG_CHECK_ERROR; 2160 break; 2161 case SPDK_DIF_GUARD_ERROR: 2162 result = SPDK_NVME_SC_GUARD_CHECK_ERROR; 2163 break; 2164 default: 2165 SPDK_UNREACHABLE(); 2166 break; 2167 } 2168 2169 return result; 2170 } 2171 2172 static void 2173 nvmf_tcp_send_c2h_data(struct spdk_nvmf_tcp_qpair *tqpair, 2174 struct spdk_nvmf_tcp_req *tcp_req) 2175 { 2176 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF( 2177 tqpair->qpair.transport, struct spdk_nvmf_tcp_transport, transport); 2178 struct nvme_tcp_pdu *rsp_pdu; 2179 struct spdk_nvme_tcp_c2h_data_hdr *c2h_data; 2180 uint32_t plen, pdo, alignment; 2181 int rc; 2182 2183 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2184 2185 rsp_pdu = nvmf_tcp_req_pdu_init(tcp_req); 2186 assert(rsp_pdu != NULL); 2187 2188 c2h_data = &rsp_pdu->hdr.c2h_data; 2189 c2h_data->common.pdu_type = SPDK_NVME_TCP_PDU_TYPE_C2H_DATA; 2190 plen = c2h_data->common.hlen = sizeof(*c2h_data); 2191 2192 if (tqpair->host_hdgst_enable) { 2193 plen += SPDK_NVME_TCP_DIGEST_LEN; 2194 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_HDGSTF; 2195 } 2196 2197 /* set the psh */ 2198 c2h_data->cccid = tcp_req->req.cmd->nvme_cmd.cid; 2199 c2h_data->datal = tcp_req->req.length; 2200 c2h_data->datao = 0; 2201 2202 /* set the padding */ 2203 rsp_pdu->padding_len = 0; 2204 pdo = plen; 2205 if (tqpair->cpda) { 2206 alignment = (tqpair->cpda + 1) << 2; 2207 if (alignment > plen) { 2208 rsp_pdu->padding_len = alignment - plen; 2209 pdo = plen = alignment; 2210 } 2211 } 2212 2213 c2h_data->common.pdo = pdo; 2214 plen += c2h_data->datal; 2215 if (tqpair->host_ddgst_enable) { 2216 c2h_data->common.flags |= SPDK_NVME_TCP_CH_FLAGS_DDGSTF; 2217 plen += SPDK_NVME_TCP_DIGEST_LEN; 2218 } 2219 2220 c2h_data->common.plen = plen; 2221 2222 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2223 rsp_pdu->dif_ctx = &tcp_req->req.dif.dif_ctx; 2224 } 2225 2226 nvme_tcp_pdu_set_data_buf(rsp_pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2227 c2h_data->datao, c2h_data->datal); 2228 2229 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2230 struct spdk_nvme_cpl *rsp = &tcp_req->req.rsp->nvme_cpl; 2231 struct spdk_dif_error err_blk = {}; 2232 2233 rc = spdk_dif_verify_stream(rsp_pdu->data_iov, rsp_pdu->data_iovcnt, 2234 0, rsp_pdu->data_len, rsp_pdu->dif_ctx, &err_blk); 2235 if (rc != 0) { 2236 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2237 err_blk.err_type, err_blk.err_offset); 2238 rsp->status.sct = SPDK_NVME_SCT_MEDIA_ERROR; 2239 rsp->status.sc = nvmf_tcp_dif_error_to_compl_status(err_blk.err_type); 2240 nvmf_tcp_req_pdu_fini(tcp_req); 2241 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2242 return; 2243 } 2244 } 2245 2246 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_LAST_PDU; 2247 if (ttransport->tcp_opts.c2h_success) { 2248 c2h_data->common.flags |= SPDK_NVME_TCP_C2H_DATA_FLAGS_SUCCESS; 2249 } 2250 2251 nvmf_tcp_qpair_write_pdu(tqpair, rsp_pdu, nvmf_tcp_pdu_c2h_data_complete, tcp_req); 2252 } 2253 2254 static int 2255 request_transfer_out(struct spdk_nvmf_request *req) 2256 { 2257 struct spdk_nvmf_tcp_req *tcp_req; 2258 struct spdk_nvmf_qpair *qpair; 2259 struct spdk_nvmf_tcp_qpair *tqpair; 2260 struct spdk_nvme_cpl *rsp; 2261 2262 SPDK_DEBUGLOG(nvmf_tcp, "enter\n"); 2263 2264 qpair = req->qpair; 2265 rsp = &req->rsp->nvme_cpl; 2266 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2267 2268 /* Advance our sq_head pointer */ 2269 if (qpair->sq_head == qpair->sq_head_max) { 2270 qpair->sq_head = 0; 2271 } else { 2272 qpair->sq_head++; 2273 } 2274 rsp->sqhd = qpair->sq_head; 2275 2276 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2277 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST); 2278 if (rsp->status.sc == SPDK_NVME_SC_SUCCESS && req->xfer == SPDK_NVME_DATA_CONTROLLER_TO_HOST) { 2279 nvmf_tcp_send_c2h_data(tqpair, tcp_req); 2280 } else { 2281 nvmf_tcp_send_capsule_resp_pdu(tcp_req, tqpair); 2282 } 2283 2284 return 0; 2285 } 2286 2287 static void 2288 nvmf_tcp_set_incapsule_data(struct spdk_nvmf_tcp_qpair *tqpair, 2289 struct spdk_nvmf_tcp_req *tcp_req) 2290 { 2291 struct nvme_tcp_pdu *pdu; 2292 uint32_t plen = 0; 2293 2294 pdu = &tqpair->pdu_in_progress; 2295 plen = pdu->hdr.common.hlen; 2296 2297 if (tqpair->host_hdgst_enable) { 2298 plen += SPDK_NVME_TCP_DIGEST_LEN; 2299 } 2300 2301 if (pdu->hdr.common.plen != plen) { 2302 tcp_req->has_incapsule_data = true; 2303 } 2304 } 2305 2306 static bool 2307 nvmf_tcp_req_process(struct spdk_nvmf_tcp_transport *ttransport, 2308 struct spdk_nvmf_tcp_req *tcp_req) 2309 { 2310 struct spdk_nvmf_tcp_qpair *tqpair; 2311 int rc; 2312 enum spdk_nvmf_tcp_req_state prev_state; 2313 bool progress = false; 2314 struct spdk_nvmf_transport *transport = &ttransport->transport; 2315 struct spdk_nvmf_transport_poll_group *group; 2316 struct spdk_nvmf_tcp_poll_group *tgroup; 2317 2318 tqpair = SPDK_CONTAINEROF(tcp_req->req.qpair, struct spdk_nvmf_tcp_qpair, qpair); 2319 group = &tqpair->group->group; 2320 assert(tcp_req->state != TCP_REQUEST_STATE_FREE); 2321 2322 /* If the qpair is not active, we need to abort the outstanding requests. */ 2323 if (tqpair->qpair.state != SPDK_NVMF_QPAIR_ACTIVE) { 2324 if (tcp_req->state == TCP_REQUEST_STATE_NEED_BUFFER) { 2325 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2326 } 2327 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_COMPLETED); 2328 } 2329 2330 /* The loop here is to allow for several back-to-back state changes. */ 2331 do { 2332 prev_state = tcp_req->state; 2333 2334 SPDK_DEBUGLOG(nvmf_tcp, "Request %p entering state %d on tqpair=%p\n", tcp_req, prev_state, 2335 tqpair); 2336 2337 switch (tcp_req->state) { 2338 case TCP_REQUEST_STATE_FREE: 2339 /* Some external code must kick a request into TCP_REQUEST_STATE_NEW 2340 * to escape this state. */ 2341 break; 2342 case TCP_REQUEST_STATE_NEW: 2343 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEW, 0, 0, (uintptr_t)tcp_req, 0); 2344 2345 /* copy the cmd from the receive pdu */ 2346 tcp_req->cmd = tqpair->pdu_in_progress.hdr.capsule_cmd.ccsqe; 2347 2348 if (spdk_unlikely(spdk_nvmf_request_get_dif_ctx(&tcp_req->req, &tcp_req->req.dif.dif_ctx))) { 2349 tcp_req->req.dif.dif_insert_or_strip = true; 2350 tqpair->pdu_in_progress.dif_ctx = &tcp_req->req.dif.dif_ctx; 2351 } 2352 2353 /* The next state transition depends on the data transfer needs of this request. */ 2354 tcp_req->req.xfer = spdk_nvmf_req_get_xfer(&tcp_req->req); 2355 2356 if (spdk_unlikely(tcp_req->req.xfer == SPDK_NVME_DATA_BIDIRECTIONAL)) { 2357 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2358 tcp_req->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SC_INVALID_OPCODE; 2359 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2360 SPDK_DEBUGLOG(nvmf_tcp, "Request %p: invalid xfer type (BIDIRECTIONAL)\n", tcp_req); 2361 break; 2362 } 2363 2364 /* If no data to transfer, ready to execute. */ 2365 if (tcp_req->req.xfer == SPDK_NVME_DATA_NONE) { 2366 /* Reset the tqpair receving pdu state */ 2367 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2368 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2369 break; 2370 } 2371 2372 nvmf_tcp_set_incapsule_data(tqpair, tcp_req); 2373 2374 if (!tcp_req->has_incapsule_data) { 2375 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_READY); 2376 } 2377 2378 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_NEED_BUFFER); 2379 STAILQ_INSERT_TAIL(&group->pending_buf_queue, &tcp_req->req, buf_link); 2380 break; 2381 case TCP_REQUEST_STATE_NEED_BUFFER: 2382 spdk_trace_record(TRACE_TCP_REQUEST_STATE_NEED_BUFFER, 0, 0, (uintptr_t)tcp_req, 0); 2383 2384 assert(tcp_req->req.xfer != SPDK_NVME_DATA_NONE); 2385 2386 if (!tcp_req->has_incapsule_data && (&tcp_req->req != STAILQ_FIRST(&group->pending_buf_queue))) { 2387 SPDK_DEBUGLOG(nvmf_tcp, 2388 "Not the first element to wait for the buf for tcp_req(%p) on tqpair=%p\n", 2389 tcp_req, tqpair); 2390 /* This request needs to wait in line to obtain a buffer */ 2391 break; 2392 } 2393 2394 /* Try to get a data buffer */ 2395 rc = nvmf_tcp_req_parse_sgl(tcp_req, transport, group); 2396 if (rc < 0) { 2397 STAILQ_REMOVE_HEAD(&group->pending_buf_queue, buf_link); 2398 /* Reset the tqpair receving pdu state */ 2399 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_ERROR); 2400 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2401 break; 2402 } 2403 2404 if (!tcp_req->req.data) { 2405 SPDK_DEBUGLOG(nvmf_tcp, "No buffer allocated for tcp_req(%p) on tqpair(%p\n)", 2406 tcp_req, tqpair); 2407 /* No buffers available. */ 2408 break; 2409 } 2410 2411 STAILQ_REMOVE(&group->pending_buf_queue, &tcp_req->req, spdk_nvmf_request, buf_link); 2412 2413 /* If data is transferring from host to controller, we need to do a transfer from the host. */ 2414 if (tcp_req->req.xfer == SPDK_NVME_DATA_HOST_TO_CONTROLLER) { 2415 if (tcp_req->req.data_from_pool) { 2416 SPDK_DEBUGLOG(nvmf_tcp, "Sending R2T for tcp_req(%p) on tqpair=%p\n", tcp_req, tqpair); 2417 nvmf_tcp_send_r2t_pdu(tqpair, tcp_req); 2418 } else { 2419 struct nvme_tcp_pdu *pdu; 2420 2421 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER); 2422 2423 pdu = &tqpair->pdu_in_progress; 2424 SPDK_DEBUGLOG(nvmf_tcp, "Not need to send r2t for tcp_req(%p) on tqpair=%p\n", tcp_req, 2425 tqpair); 2426 /* No need to send r2t, contained in the capsuled data */ 2427 nvme_tcp_pdu_set_data_buf(pdu, tcp_req->req.iov, tcp_req->req.iovcnt, 2428 0, tcp_req->req.length); 2429 nvmf_tcp_qpair_set_recv_state(tqpair, NVME_TCP_PDU_RECV_STATE_AWAIT_PDU_PAYLOAD); 2430 } 2431 break; 2432 } 2433 2434 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_EXECUTE); 2435 break; 2436 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2437 spdk_trace_record(TRACE_TCP_REQUEST_STATE_AWAIT_R2T_ACK, 0, 0, (uintptr_t)tcp_req, 0); 2438 /* The R2T completion or the h2c data incoming will kick it out of this state. */ 2439 break; 2440 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2441 2442 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER, 0, 0, 2443 (uintptr_t)tcp_req, 0); 2444 /* Some external code must kick a request into TCP_REQUEST_STATE_READY_TO_EXECUTE 2445 * to escape this state. */ 2446 break; 2447 case TCP_REQUEST_STATE_READY_TO_EXECUTE: 2448 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_EXECUTE, 0, 0, (uintptr_t)tcp_req, 0); 2449 2450 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2451 assert(tcp_req->req.dif.elba_length >= tcp_req->req.length); 2452 tcp_req->req.length = tcp_req->req.dif.elba_length; 2453 } 2454 2455 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTING); 2456 spdk_nvmf_request_exec(&tcp_req->req); 2457 break; 2458 case TCP_REQUEST_STATE_EXECUTING: 2459 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTING, 0, 0, (uintptr_t)tcp_req, 0); 2460 /* Some external code must kick a request into TCP_REQUEST_STATE_EXECUTED 2461 * to escape this state. */ 2462 break; 2463 case TCP_REQUEST_STATE_EXECUTED: 2464 spdk_trace_record(TRACE_TCP_REQUEST_STATE_EXECUTED, 0, 0, (uintptr_t)tcp_req, 0); 2465 2466 if (spdk_unlikely(tcp_req->req.dif.dif_insert_or_strip)) { 2467 tcp_req->req.length = tcp_req->req.dif.orig_length; 2468 } 2469 2470 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2471 break; 2472 case TCP_REQUEST_STATE_READY_TO_COMPLETE: 2473 spdk_trace_record(TRACE_TCP_REQUEST_STATE_READY_TO_COMPLETE, 0, 0, (uintptr_t)tcp_req, 0); 2474 rc = request_transfer_out(&tcp_req->req); 2475 assert(rc == 0); /* No good way to handle this currently */ 2476 break; 2477 case TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST: 2478 spdk_trace_record(TRACE_TCP_REQUEST_STATE_TRANSFERRING_CONTROLLER_TO_HOST, 0, 0, 2479 (uintptr_t)tcp_req, 2480 0); 2481 /* Some external code must kick a request into TCP_REQUEST_STATE_COMPLETED 2482 * to escape this state. */ 2483 break; 2484 case TCP_REQUEST_STATE_COMPLETED: 2485 spdk_trace_record(TRACE_TCP_REQUEST_STATE_COMPLETED, 0, 0, (uintptr_t)tcp_req, 0); 2486 if (tcp_req->req.data_from_pool) { 2487 spdk_nvmf_request_free_buffers(&tcp_req->req, group, transport); 2488 } else if (spdk_unlikely(tcp_req->has_incapsule_data && (tcp_req->cmd.opc == SPDK_NVME_OPC_FABRIC || 2489 tqpair->qpair.qid == 0) && tcp_req->req.length > transport->opts.in_capsule_data_size)) { 2490 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2491 assert(tgroup->control_msg_list); 2492 SPDK_DEBUGLOG(nvmf_tcp, "Put buf to control msg list\n"); 2493 nvmf_tcp_control_msg_put(tgroup->control_msg_list, tcp_req->req.data); 2494 } 2495 tcp_req->req.length = 0; 2496 tcp_req->req.iovcnt = 0; 2497 tcp_req->req.data = NULL; 2498 2499 nvmf_tcp_req_pdu_fini(tcp_req); 2500 2501 nvmf_tcp_req_put(tqpair, tcp_req); 2502 break; 2503 case TCP_REQUEST_NUM_STATES: 2504 default: 2505 assert(0); 2506 break; 2507 } 2508 2509 if (tcp_req->state != prev_state) { 2510 progress = true; 2511 } 2512 } while (tcp_req->state != prev_state); 2513 2514 return progress; 2515 } 2516 2517 static void 2518 nvmf_tcp_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock) 2519 { 2520 struct spdk_nvmf_tcp_qpair *tqpair = arg; 2521 int rc; 2522 2523 assert(tqpair != NULL); 2524 rc = nvmf_tcp_sock_process(tqpair); 2525 2526 /* If there was a new socket error, disconnect */ 2527 if (rc < 0) { 2528 nvmf_tcp_qpair_disconnect(tqpair); 2529 } 2530 } 2531 2532 static int 2533 nvmf_tcp_poll_group_add(struct spdk_nvmf_transport_poll_group *group, 2534 struct spdk_nvmf_qpair *qpair) 2535 { 2536 struct spdk_nvmf_tcp_poll_group *tgroup; 2537 struct spdk_nvmf_tcp_qpair *tqpair; 2538 int rc; 2539 2540 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2541 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2542 2543 rc = spdk_sock_group_add_sock(tgroup->sock_group, tqpair->sock, 2544 nvmf_tcp_sock_cb, tqpair); 2545 if (rc != 0) { 2546 SPDK_ERRLOG("Could not add sock to sock_group: %s (%d)\n", 2547 spdk_strerror(errno), errno); 2548 return -1; 2549 } 2550 2551 rc = nvmf_tcp_qpair_sock_init(tqpair); 2552 if (rc != 0) { 2553 SPDK_ERRLOG("Cannot set sock opt for tqpair=%p\n", tqpair); 2554 return -1; 2555 } 2556 2557 rc = nvmf_tcp_qpair_init(&tqpair->qpair); 2558 if (rc < 0) { 2559 SPDK_ERRLOG("Cannot init tqpair=%p\n", tqpair); 2560 return -1; 2561 } 2562 2563 rc = nvmf_tcp_qpair_init_mem_resource(tqpair); 2564 if (rc < 0) { 2565 SPDK_ERRLOG("Cannot init memory resource info for tqpair=%p\n", tqpair); 2566 return -1; 2567 } 2568 2569 tqpair->group = tgroup; 2570 tqpair->state = NVME_TCP_QPAIR_STATE_INVALID; 2571 TAILQ_INSERT_TAIL(&tgroup->qpairs, tqpair, link); 2572 2573 return 0; 2574 } 2575 2576 static int 2577 nvmf_tcp_poll_group_remove(struct spdk_nvmf_transport_poll_group *group, 2578 struct spdk_nvmf_qpair *qpair) 2579 { 2580 struct spdk_nvmf_tcp_poll_group *tgroup; 2581 struct spdk_nvmf_tcp_qpair *tqpair; 2582 int rc; 2583 2584 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2585 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2586 2587 assert(tqpair->group == tgroup); 2588 2589 SPDK_DEBUGLOG(nvmf_tcp, "remove tqpair=%p from the tgroup=%p\n", tqpair, tgroup); 2590 if (tqpair->recv_state == NVME_TCP_PDU_RECV_STATE_AWAIT_REQ) { 2591 TAILQ_REMOVE(&tgroup->await_req, tqpair, link); 2592 } else { 2593 TAILQ_REMOVE(&tgroup->qpairs, tqpair, link); 2594 } 2595 2596 rc = spdk_sock_group_remove_sock(tgroup->sock_group, tqpair->sock); 2597 if (rc != 0) { 2598 SPDK_ERRLOG("Could not remove sock from sock_group: %s (%d)\n", 2599 spdk_strerror(errno), errno); 2600 } 2601 2602 return rc; 2603 } 2604 2605 static int 2606 nvmf_tcp_req_complete(struct spdk_nvmf_request *req) 2607 { 2608 struct spdk_nvmf_tcp_transport *ttransport; 2609 struct spdk_nvmf_tcp_req *tcp_req; 2610 2611 ttransport = SPDK_CONTAINEROF(req->qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2612 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2613 2614 nvmf_tcp_req_set_state(tcp_req, TCP_REQUEST_STATE_EXECUTED); 2615 nvmf_tcp_req_process(ttransport, tcp_req); 2616 2617 return 0; 2618 } 2619 2620 static void 2621 nvmf_tcp_close_qpair(struct spdk_nvmf_qpair *qpair, 2622 spdk_nvmf_transport_qpair_fini_cb cb_fn, void *cb_arg) 2623 { 2624 struct spdk_nvmf_tcp_qpair *tqpair; 2625 2626 SPDK_DEBUGLOG(nvmf_tcp, "Qpair: %p\n", qpair); 2627 2628 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2629 tqpair->state = NVME_TCP_QPAIR_STATE_EXITED; 2630 nvmf_tcp_qpair_destroy(tqpair); 2631 2632 if (cb_fn) { 2633 cb_fn(cb_arg); 2634 } 2635 } 2636 2637 static int 2638 nvmf_tcp_poll_group_poll(struct spdk_nvmf_transport_poll_group *group) 2639 { 2640 struct spdk_nvmf_tcp_poll_group *tgroup; 2641 int rc; 2642 struct spdk_nvmf_request *req, *req_tmp; 2643 struct spdk_nvmf_tcp_req *tcp_req; 2644 struct spdk_nvmf_tcp_qpair *tqpair, *tqpair_tmp; 2645 struct spdk_nvmf_tcp_transport *ttransport = SPDK_CONTAINEROF(group->transport, 2646 struct spdk_nvmf_tcp_transport, transport); 2647 2648 tgroup = SPDK_CONTAINEROF(group, struct spdk_nvmf_tcp_poll_group, group); 2649 2650 if (spdk_unlikely(TAILQ_EMPTY(&tgroup->qpairs) && TAILQ_EMPTY(&tgroup->await_req))) { 2651 return 0; 2652 } 2653 2654 STAILQ_FOREACH_SAFE(req, &group->pending_buf_queue, buf_link, req_tmp) { 2655 tcp_req = SPDK_CONTAINEROF(req, struct spdk_nvmf_tcp_req, req); 2656 if (nvmf_tcp_req_process(ttransport, tcp_req) == false) { 2657 break; 2658 } 2659 } 2660 2661 rc = spdk_sock_group_poll(tgroup->sock_group); 2662 if (rc < 0) { 2663 SPDK_ERRLOG("Failed to poll sock_group=%p\n", tgroup->sock_group); 2664 } 2665 2666 TAILQ_FOREACH_SAFE(tqpair, &tgroup->await_req, link, tqpair_tmp) { 2667 nvmf_tcp_sock_process(tqpair); 2668 } 2669 2670 return rc; 2671 } 2672 2673 static int 2674 nvmf_tcp_qpair_get_trid(struct spdk_nvmf_qpair *qpair, 2675 struct spdk_nvme_transport_id *trid, bool peer) 2676 { 2677 struct spdk_nvmf_tcp_qpair *tqpair; 2678 uint16_t port; 2679 2680 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2681 spdk_nvme_trid_populate_transport(trid, SPDK_NVME_TRANSPORT_TCP); 2682 2683 if (peer) { 2684 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->initiator_addr); 2685 port = tqpair->initiator_port; 2686 } else { 2687 snprintf(trid->traddr, sizeof(trid->traddr), "%s", tqpair->target_addr); 2688 port = tqpair->target_port; 2689 } 2690 2691 if (spdk_sock_is_ipv4(tqpair->sock)) { 2692 trid->adrfam = SPDK_NVMF_ADRFAM_IPV4; 2693 } else if (spdk_sock_is_ipv6(tqpair->sock)) { 2694 trid->adrfam = SPDK_NVMF_ADRFAM_IPV6; 2695 } else { 2696 return -1; 2697 } 2698 2699 snprintf(trid->trsvcid, sizeof(trid->trsvcid), "%d", port); 2700 return 0; 2701 } 2702 2703 static int 2704 nvmf_tcp_qpair_get_local_trid(struct spdk_nvmf_qpair *qpair, 2705 struct spdk_nvme_transport_id *trid) 2706 { 2707 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2708 } 2709 2710 static int 2711 nvmf_tcp_qpair_get_peer_trid(struct spdk_nvmf_qpair *qpair, 2712 struct spdk_nvme_transport_id *trid) 2713 { 2714 return nvmf_tcp_qpair_get_trid(qpair, trid, 1); 2715 } 2716 2717 static int 2718 nvmf_tcp_qpair_get_listen_trid(struct spdk_nvmf_qpair *qpair, 2719 struct spdk_nvme_transport_id *trid) 2720 { 2721 return nvmf_tcp_qpair_get_trid(qpair, trid, 0); 2722 } 2723 2724 static void 2725 nvmf_tcp_req_set_abort_status(struct spdk_nvmf_request *req, 2726 struct spdk_nvmf_tcp_req *tcp_req_to_abort) 2727 { 2728 tcp_req_to_abort->req.rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 2729 tcp_req_to_abort->req.rsp->nvme_cpl.status.sc = SPDK_NVME_SC_ABORTED_BY_REQUEST; 2730 2731 nvmf_tcp_req_set_state(tcp_req_to_abort, TCP_REQUEST_STATE_READY_TO_COMPLETE); 2732 2733 req->rsp->nvme_cpl.cdw0 &= ~1U; /* Command was successfully aborted. */ 2734 } 2735 2736 static int 2737 _nvmf_tcp_qpair_abort_request(void *ctx) 2738 { 2739 struct spdk_nvmf_request *req = ctx; 2740 struct spdk_nvmf_tcp_req *tcp_req_to_abort = SPDK_CONTAINEROF(req->req_to_abort, 2741 struct spdk_nvmf_tcp_req, req); 2742 struct spdk_nvmf_tcp_qpair *tqpair = SPDK_CONTAINEROF(req->req_to_abort->qpair, 2743 struct spdk_nvmf_tcp_qpair, qpair); 2744 int rc; 2745 2746 spdk_poller_unregister(&req->poller); 2747 2748 switch (tcp_req_to_abort->state) { 2749 case TCP_REQUEST_STATE_EXECUTING: 2750 rc = nvmf_ctrlr_abort_request(req); 2751 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS) { 2752 return SPDK_POLLER_BUSY; 2753 } 2754 break; 2755 2756 case TCP_REQUEST_STATE_NEED_BUFFER: 2757 STAILQ_REMOVE(&tqpair->group->group.pending_buf_queue, 2758 &tcp_req_to_abort->req, spdk_nvmf_request, buf_link); 2759 2760 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2761 break; 2762 2763 case TCP_REQUEST_STATE_AWAITING_R2T_ACK: 2764 nvmf_tcp_req_set_abort_status(req, tcp_req_to_abort); 2765 break; 2766 2767 case TCP_REQUEST_STATE_TRANSFERRING_HOST_TO_CONTROLLER: 2768 if (spdk_get_ticks() < req->timeout_tsc) { 2769 req->poller = SPDK_POLLER_REGISTER(_nvmf_tcp_qpair_abort_request, req, 0); 2770 return SPDK_POLLER_BUSY; 2771 } 2772 break; 2773 2774 default: 2775 break; 2776 } 2777 2778 spdk_nvmf_request_complete(req); 2779 return SPDK_POLLER_BUSY; 2780 } 2781 2782 static void 2783 nvmf_tcp_qpair_abort_request(struct spdk_nvmf_qpair *qpair, 2784 struct spdk_nvmf_request *req) 2785 { 2786 struct spdk_nvmf_tcp_qpair *tqpair; 2787 struct spdk_nvmf_tcp_transport *ttransport; 2788 struct spdk_nvmf_transport *transport; 2789 uint16_t cid; 2790 uint32_t i; 2791 struct spdk_nvmf_tcp_req *tcp_req_to_abort = NULL; 2792 2793 tqpair = SPDK_CONTAINEROF(qpair, struct spdk_nvmf_tcp_qpair, qpair); 2794 ttransport = SPDK_CONTAINEROF(qpair->transport, struct spdk_nvmf_tcp_transport, transport); 2795 transport = &ttransport->transport; 2796 2797 cid = req->cmd->nvme_cmd.cdw10_bits.abort.cid; 2798 2799 for (i = 0; i < tqpair->resource_count; i++) { 2800 if (tqpair->reqs[i].state != TCP_REQUEST_STATE_FREE && 2801 tqpair->reqs[i].req.cmd->nvme_cmd.cid == cid) { 2802 tcp_req_to_abort = &tqpair->reqs[i]; 2803 break; 2804 } 2805 } 2806 2807 if (tcp_req_to_abort == NULL) { 2808 spdk_nvmf_request_complete(req); 2809 return; 2810 } 2811 2812 req->req_to_abort = &tcp_req_to_abort->req; 2813 req->timeout_tsc = spdk_get_ticks() + 2814 transport->opts.abort_timeout_sec * spdk_get_ticks_hz(); 2815 req->poller = NULL; 2816 2817 _nvmf_tcp_qpair_abort_request(req); 2818 } 2819 2820 #define SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH 128 2821 #define SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH 128 2822 #define SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR 128 2823 #define SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE 4096 2824 #define SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE 131072 2825 #define SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE 131072 2826 #define SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS 511 2827 #define SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE 32 2828 #define SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP false 2829 #define SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC 1 2830 2831 static void 2832 nvmf_tcp_opts_init(struct spdk_nvmf_transport_opts *opts) 2833 { 2834 opts->max_queue_depth = SPDK_NVMF_TCP_DEFAULT_MAX_QUEUE_DEPTH; 2835 opts->max_qpairs_per_ctrlr = SPDK_NVMF_TCP_DEFAULT_MAX_QPAIRS_PER_CTRLR; 2836 opts->in_capsule_data_size = SPDK_NVMF_TCP_DEFAULT_IN_CAPSULE_DATA_SIZE; 2837 opts->max_io_size = SPDK_NVMF_TCP_DEFAULT_MAX_IO_SIZE; 2838 opts->io_unit_size = SPDK_NVMF_TCP_DEFAULT_IO_UNIT_SIZE; 2839 opts->max_aq_depth = SPDK_NVMF_TCP_DEFAULT_AQ_DEPTH; 2840 opts->num_shared_buffers = SPDK_NVMF_TCP_DEFAULT_NUM_SHARED_BUFFERS; 2841 opts->buf_cache_size = SPDK_NVMF_TCP_DEFAULT_BUFFER_CACHE_SIZE; 2842 opts->dif_insert_or_strip = SPDK_NVMF_TCP_DEFAULT_DIF_INSERT_OR_STRIP; 2843 opts->abort_timeout_sec = SPDK_NVMF_TCP_DEFAULT_ABORT_TIMEOUT_SEC; 2844 opts->transport_specific = NULL; 2845 } 2846 2847 const struct spdk_nvmf_transport_ops spdk_nvmf_transport_tcp = { 2848 .name = "TCP", 2849 .type = SPDK_NVME_TRANSPORT_TCP, 2850 .opts_init = nvmf_tcp_opts_init, 2851 .create = nvmf_tcp_create, 2852 .dump_opts = nvmf_tcp_dump_opts, 2853 .destroy = nvmf_tcp_destroy, 2854 2855 .listen = nvmf_tcp_listen, 2856 .stop_listen = nvmf_tcp_stop_listen, 2857 .accept = nvmf_tcp_accept, 2858 2859 .listener_discover = nvmf_tcp_discover, 2860 2861 .poll_group_create = nvmf_tcp_poll_group_create, 2862 .get_optimal_poll_group = nvmf_tcp_get_optimal_poll_group, 2863 .poll_group_destroy = nvmf_tcp_poll_group_destroy, 2864 .poll_group_add = nvmf_tcp_poll_group_add, 2865 .poll_group_remove = nvmf_tcp_poll_group_remove, 2866 .poll_group_poll = nvmf_tcp_poll_group_poll, 2867 2868 .req_free = nvmf_tcp_req_free, 2869 .req_complete = nvmf_tcp_req_complete, 2870 2871 .qpair_fini = nvmf_tcp_close_qpair, 2872 .qpair_get_local_trid = nvmf_tcp_qpair_get_local_trid, 2873 .qpair_get_peer_trid = nvmf_tcp_qpair_get_peer_trid, 2874 .qpair_get_listen_trid = nvmf_tcp_qpair_get_listen_trid, 2875 .qpair_abort_request = nvmf_tcp_qpair_abort_request, 2876 }; 2877 2878 SPDK_NVMF_TRANSPORT_REGISTER(tcp, &spdk_nvmf_transport_tcp); 2879 SPDK_LOG_REGISTER_COMPONENT(nvmf_tcp) 2880